[
{
"n": 1,
"year": 2025,
"title": "Car Damage Detection using YOLOv11 and RT-DETR",
"authors": "Azmi Rizqullah Rabbani, Gamma Kosala",
"venue": "International Seminar on Research of Information Technology and Intelligent Systems (ISRITI)",
"cites": 0,
"link": "https://doi.org/10.1109/isriti68345.2025.11393155",
"tasks": "detection",
"models": "YOLOv11; YOLOv10; RT-DETR; DETR",
"datasets": "CarDD",
"reported_metrics": "accuracy 63.34%",
"techniques": "lightweight / mobile deployment; transformer backbone",
"seed": true,
"abstract": "Recent progress in deep learning and computer vision has enabled the automation of car damage detection using object detection techniques. This study conducts a comparative evaluation of three cutting-edge detection models: YOLOv11m, YOLOv10m, and RT-DETR-L on the Car Damage Detection (CarDD) dataset, focusing on six distinct types of vehicle damage. The main objective is to compare the newest YOLOv11m architecture with the transformer-based RT-DETR-L, while YOLOv10m serves as a robust convolutional baseline to highlight performance differences. All models are trained under the same conditions using a stable-tuned training strategy that emphasizes training consistency and validation reliability over aggressive optimization. The models are assessed based on mean Average Precision at IoU 0.5 (mAP@50), mean Average Precision across IoU thresholds from 0.5 to 0.95 (mAP@50:95), Precision, Recall, and inference time. The CarDD dataset includes 4,000 annotated images across six damage categories (dent, scratch, crack, glass shatter, lamp broken, and tire flat), allowing for a comprehensive class-wise evaluation. Experimental results show that YOLOv11m delivers the highest overall accuracy (mAP@50:95: 63.34%) with precise detection of minor and fine-grained damages like scratches and cracks. YOLOv10m strikes a strong balance between performance and speed (61.6 FPS), making it a practical option for real-time applications. RT-DETR-L shows strong Precision and Recall, supported by its "
},
{
"n": 2,
"year": 2026,
"title": "Grounding Agentic VLMs with Dedicated Segmentation for Fine-Grained Vehicle Damage Assessment",
"authors": "Vishwajeet Shivaji Hogale, Anjali Pai, Nitya Ravi",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2608.02470",
"tasks": "segmentation; detection; classification; VLM",
"models": "Qwen-VL; VLM/MLLM",
"datasets": "",
"reported_metrics": "accuracy 87.3%",
"techniques": "vision-language / LLM; grounding / text-guided; reflection / false-positive handling; loss redesign (focal/IoU variants)",
"seed": true,
"abstract": "Vision-language models (VLMs) are increasingly deployed as reasoning agents in real-world visual assessment pipelines, yet their spatial grounding remains unreliable for fine-grained, visually ambiguous targets. We study this gap in the context of automated vehicle damage assessment, where fine-grained defects such as scratches and hairline cracks occupy few pixels, produce weak gradient signal, and are easily confused with reflections and surface texture. We show that a state-of-the-art VLM (Qwen-VL) achieves strong semantic classification accuracy (87.3%) on this task but is systematically ungrounded at the spatial level: it hallucinates damage in reflective regions, misses elongated scratches entirely, and produces spatially inconsistent outputs when prompted for localization. We propose TinyDamage, a hybrid architecture that delegates spatial grounding to a dedicated multi-task segmentation model while reserving the VLM for semantic reasoning and report generation. On the segmentation side, we find that the choice of loss function has an outsized and underexplored effect on tiny-object grounding: focal loss, widely used for class imbalance, collapses tiny-damage detection to zero, while a supervised contrastive objective measurably improves damage/background separability. We integrate the segmentation model into a 7-node LangGraph agent pipeline that grounds every VLM generation step in the segmentation output, and show that this grounding reduces the report hallucination"
},
{
"n": 3,
"year": 2025,
"title": "Smart Car Damage Assessment Using Enhanced YOLO Algorithm and Image Processing Techniques",
"authors": "Muhammad Remzy Syah Ramazhan, Alhadi Bustamam, Rinaldi Anwar Buyung",
"venue": "Information",
"cites": 12,
"link": "https://doi.org/10.3390/info16030211",
"tasks": "detection; severity",
"models": "YOLOv9",
"datasets": "CarDD",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); severity / cost estimation; loss redesign (focal/IoU variants)",
"seed": true,
"abstract": "Conventional inspections in car damage assessments depend on visual judgments by human inspectors, which are labor-intensive and prone to fraudulent practices through manipulating damages. Recent advancements in artificial intelligence have given rise to a state-of-the-art object detection algorithm, the You Only Look Once algorithm (YOLO), that sets a new standard in smart and automated damage assessment. This study proposes an enhanced YOLOv9 network tailored to detect six types of car damage. The enhancements include the convolutional block attention module (CBAM), applied to the backbone layer to enhance the model’s ability to focus on key damaged regions, and the SCYLLA-IoU (SIoU) loss function, introduced for bounding box regression. To be able to assess the damage severity comprehensively, we propose a novel formula named damage severity index (DSI) for quantifying damage severity directly from images, integrating multiple factors such as the number of detected damages, the ratio of damage to the image size, object detection confidence, and the type of damage. Experimental results on the CarDD dataset show that the proposed model outperforms state-of-the-art YOLO algorithms by 1.75% and that the proposed DSI demonstrates intuitive assessment of damage severity with numbers, aiding repair decisions."
},
{
"n": 4,
"year": 2025,
"title": "Improved YOLOv11 algorithm for vehicle damage detection",
"authors": "Changming Zhu, Jie Ji",
"venue": "",
"cites": 0,
"link": "https://doi.org/10.1117/12.3071544",
"tasks": "detection; severity",
"models": "YOLOv11; RT-DETR; DETR",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate)",
"seed": true,
"abstract": "With the acceleration of global economic development and urbanization, the number of vehicles has been increasing year by year, especially in congested urban environments, where the risk of vehicle damage has increased significantly. Vehicle damage detection has become the key to auto insurance claims, intelligent vehicle management and maintenance efficiency improvement. Traditional manual detection methods have low efficiency and high cost, so it is hard to satisfy the requirement of intelligence management. To this end, this research introduces a car damage detection method based on an improved YOLOv11 algorithm, called CAA-YOLOv11, on the CarDamageDetection datasets using three key improvements: first, the Context Guided Block module (CG block) is used to replace the traditional convolutional layer to elevate the feature extraction competence; second, the AIFI module from RT-DETR is incorporated to replace the SPPF component in YOLOv11, thereby enhancing the model's detection capabilities in complex scenarios through the application of the self-attention mechanism; finally, the ASFFHead module is integrated into YOLOv11 to optimize the detection head and improve the scale invariance and detection accuracy by adaptive fusion of features at different scales. Experimental findings demonstrate that CAA-YOLOv11 substantially boosts the overall performance of YOLOv11, which provides strong technical support for intelligent vehicle management and automated auto insurance claims."
},
{
"n": 5,
"year": 2026,
"title": "CarDamageEval: Benchmark Evaluation of Car Damage Assessment Using Vision Language Models",
"authors": "Md Jahid Hasan, Cong Kha Nguyen, Yee Ling Boo, Hamed Jahani",
"venue": "Communications in computer and information science",
"cites": 0,
"link": "https://doi.org/10.1007/978-981-95-6786-7_17",
"tasks": "VLM",
"models": "VLM/MLLM",
"datasets": "",
"reported_metrics": "",
"techniques": "vision-language / LLM",
"seed": true,
"abstract": ""
},
{
"n": 6,
"year": 2025,
"title": "Damage Detection Using YOLOv8 AI for Vehicle Assessment",
"authors": "B. Vanathi, R Akshaya, P. Alfina., Gayathri. V.",
"venue": "",
"cites": 0,
"link": "https://doi.org/10.38124/ijisrt/25mar567",
"tasks": "detection",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": true,
"abstract": "Vehicle damage detection is an essential task in automotive assessment, insurance claim processing, and fleet management. Traditional methods involve manual inspection, which is time-consuming and prone to errors. This paper presents an automated damage detection approach utilizing YOLOv8 (You Only Look Once version 8), a state-of-the-art deep learning model for object detection. Our methodology involves training the model on a dataset comprising images of vehicles with and without damage, using supervised learning techniques. The model achieves high detection accuracy and efficiency, making it suitable for real-world applications. This study compares YOLOv8 with previous versions and alternative models to highlight improvements in speed and precision. The findings suggest that this approach can significantly enhance vehicle assessment processes, reducing human effort and improving consistency in damage evaluation."
},
{
"n": 7,
"year": 2025,
"title": "Car Damage Detection Based on Multi-View Fusion and Alignment: Dataset and Method",
"authors": "Jinbo Peng, Shoubin Dong, Hua Yuan, Xiaorou Zheng",
"venue": "IEEE Transactions on Intelligent Transportation Systems",
"cites": 3,
"link": "https://doi.org/10.1109/tits.2025.3542174",
"tasks": "detection; severity; multi-view/3D",
"models": "",
"datasets": "CDDM",
"reported_metrics": "",
"techniques": "multi-view / multi-image fusion; severity / cost estimation",
"seed": true,
"abstract": "Traffic accidents remain a significant concern due to their potential severity and impact on society. The rapid and accurate detection of car damage is increasingly crucial. Manual assessment of car damage usually relies on multi-view car images taken at the scene, which can provide richer information for damage assessment. However, most car damage algorithms are based on single-view datasets, and it is hard to fully leverage the complementary information and alignment information between distant view and close-up images. In this paper, we propose the Multi-View Car Damage Detection model (MVA-CDD), comprising three key modules: Feature Split (FS), Feature Fusion (FF), and Image Alignment (IA). The FS module extracts global and detailed information from distant-view and close-up images separately, which are then combined by the FF module. The IA module effectively aligns car damage information in distant-view and close-up images to correct errors and biases. Meanwhile, we created the new Car Damage Detection Multi-view dataset (CDDM), which has a significant advantage in both image quantity and diversity across categories, addressing the shortcomings of existing multi-view datasets. Our proposed MVA-CDD outperforms the state-of-the-art single-view and multi-view models with the dataset. Results from ablation studies further confirm the efficiency of MVA-CDD. This study contributes to optimizing the car damage detection and claims adjudication process, leading to significant l"
},
{
"n": 8,
"year": 2025,
"title": "Vehicle Damage Detection Using Artificial Intelligence: A Systematic Literature Review",
"authors": "Md Jahid Hasan, Cong Kha Nguyen, Yee Ling Boo, Hamed Jahani",
"venue": "Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery",
"cites": 11,
"link": "https://doi.org/10.1002/widm.70027",
"tasks": "detection; severity",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "severity / cost estimation",
"seed": true,
"abstract": "ABSTRACT Automating vehicle damage detection is essential for automotive industry applications like insurance claims, online sales, and repair cost estimates, addressing the labor‐intensive, time‐consuming, and error‐prone nature of current manual inspections. This systematic literature review explores the use of artificial intelligence (AI), particularly deep learning‐based algorithms, to improve the accuracy and efficiency of damage detection under dynamic and challenging conditions specific to the requirements of our industry partners. The review is structured around five key research questions and includes extensive empirical evaluations to identify gaps and challenges in existing methods. Findings reveal significant potential for AI to automate and enhance the damage detection process but also highlight areas requiring further research and development. The review discusses these gaps in detail, providing a comprehensive foundation for future work in this field. Furthermore, the review findings are intended to guide both our research and the broader research community in advancing the practical application of AI for vehicle damage assessment. The insights gained from this review are crucial for developing robust AI solutions that can operate effectively in real‐world scenarios, ultimately improving operational efficiency and customer experience in the automotive industry."
},
{
"n": 9,
"year": 2022,
"title": "CarDD: A New Dataset for Vision-based Car Damage Detection",
"authors": "Xinkuang Wang, Wenjing Li, Zhongcheng Wu",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2211.00945",
"tasks": "segmentation; detection",
"models": "",
"datasets": "CarDD",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); dataset contribution",
"seed": true,
"abstract": "Automatic car damage detection has attracted significant attention in the car insurance business. However, due to the lack of high-quality and publicly available datasets, we can hardly learn a feasible model for car damage detection. To this end, we contribute with Car Damage Detection (CarDD), the first public large-scale dataset designed for vision-based car damage detection and segmentation. Our CarDD contains 4,000 highresolution car damage images with over 9,000 well-annotated instances of six damage categories. We detail the image collection, selection, and annotation processes, and present a statistical dataset analysis. Furthermore, we conduct extensive experiments on CarDD with state-of-the-art deep methods for different tasks and provide comprehensive analyses to highlight the specialty of car damage detection. CarDD dataset and the source code are available at https://cardd-ustc.github.io."
},
{
"n": 10,
"year": 2024,
"title": "INS-MMBench: A Comprehensive Benchmark for Evaluating LVLMs' Performance in Insurance",
"authors": "Chenwei Lin, Hanjia Lyu, Xian Xu, Jiebo Luo",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2406.09105",
"tasks": "VLM",
"models": "GPT-4V/4o; LLaVA; VLM/MLLM",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "vision-language / LLM",
"seed": true,
"abstract": "Large Vision-Language Models (LVLMs) and Multimodal Large Language Models (MLLMs) have demonstrated outstanding performance in various general multimodal applications and have shown increasing promise in specialized domains. However, their potential in the insurance domain-characterized by diverse application scenarios and rich multimodal data-remains largely underexplored. To date, there is no systematic review of multimodal tasks, nor a benchmark specifically designed to assess the capabilities of LVLMs in insurance. This gap hinders the development of LVLMs within the insurance industry. This study systematically reviews and categorizes multimodal tasks for 4 representative types of insurance: auto, property, health, and agricultural. We introduce INS-MMBench, the first hierarchical benchmark tailored for the insurance domain. INS-MMBench encompasses 22 fundamental tasks, 12 meta-tasks and 5 scenario tasks, enabling a comprehensive and progressive assessment from basic capabilities to real-world use cases. We benchmark 11 leading LVLMs, including closed-source models such as GPT-4o and open-source models like LLaVA. Our evaluation validates the effectiveness of INS-MMBench and offers detailed insights into the strengths and limitations of current LVLMs on a variety of insurance-related multimodal tasks. We hope that INS-MMBench will accelerate the integration of LVLMs into the insurance industry and foster interdisciplinary research. Our dataset and evaluation code are ava"
},
{
"n": 11,
"year": 2025,
"title": "Small Dents, Big Impact: A Dataset and Deep Learning Approach for Vehicle Dent Detection",
"authors": "Danish Zia Baig, Mohsin Kamal, Zahid Ullah",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2508.15431v2",
"tasks": "detection",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "",
"techniques": "data augmentation; lightweight / mobile deployment",
"seed": true,
"abstract": "Conventional car damage inspection techniques are labor-intensive, manual, and frequently overlook tiny surface imperfections like microscopic dents. Machine learning provides an innovative solution to the increasing demand for quicker and more precise inspection methods. The paper uses the YOLOv8 object recognition framework to provide a deep learning-based solution for automatically detecting microscopic surface flaws, notably tiny dents, on car exteriors. Traditional automotive damage inspection procedures are manual, time-consuming, and frequently unreliable at detecting tiny flaws. To solve this, a bespoke dataset containing annotated photos of car surfaces under various lighting circumstances, angles, and textures was created. To improve robustness, the YOLOv8m model and its customized variants, YOLOv8m-t4 and YOLOv8m-t42, were trained employing real-time data augmentation approaches. Experimental results show that the technique has excellent detection accuracy and low inference latency, making it suited for real-time applications such as automated insurance evaluations and automobile inspections. Evaluation parameters such as mean Average Precision (mAP), precision, recall, and F1-score verified the model's efficacy. With a precision of 0.86, recall of 0.84, and F1-score of 0.85, the YOLOv8m-t42 model outperformed the YOLOv8m-t4 model (precision: 0.81, recall: 0.79, F1-score: 0.80) in identifying microscopic surface defects. With a little reduced mAP@0.5:0.95 of 0.20, "
},
{
"n": 12,
"year": 2025,
"title": "RF-DETR: Neural Architecture Search for Real-Time Detection Transformers",
"authors": "Isaac Robinson, Peter Robicheaux, Matvei Popov, Deva Ramanan",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2511.09554",
"tasks": "detection; VLM",
"models": "DETR; VLM/MLLM; Transfer learning",
"datasets": "COCO-car-damage; Roboflow",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; lightweight / mobile deployment; vision-language / LLM; grounding / text-guided; transformer backbone",
"seed": true,
"abstract": "Open-vocabulary detectors achieve impressive performance on COCO, but often fail to generalize to real-world datasets with out-of-distribution classes not typically found in their pre-training. Rather than simply fine-tuning a heavy-weight vision-language model (VLM) for new domains, we introduce RF-DETR, a light-weight specialist detection transformer that discovers accuracy-latency Pareto curves for any target dataset with weight-sharing neural architecture search (NAS). Our approach fine-tunes a pre-trained base network on a target dataset and evaluates thousands of network configurations with different accuracy-latency tradeoffs without re-training. Further, we revisit the \"tunable knobs\" for NAS to improve the transferability of DETRs to diverse target domains. Notably, RF-DETR significantly improves over prior state-of-the-art real-time methods on COCO and Roboflow100-VL. RF-DETR (nano) achieves 48.0 AP on COCO, beating D-FINE (nano) by 5.3 AP at similar latency, and RF-DETR (2x-large) outperforms GroundingDINO (tiny) by 1.2 AP on Roboflow100-VL while running 20x as fast. To the best of our knowledge, RF-DETR (2x-large) is the first real-time detector to surpass 60 AP on COCO. Our code is available at https://github.com/roboflow/rf-detr"
},
{
"n": 13,
"year": 2023,
"title": "Detects Damage Car Body using YOLO Deep Learning Algorithm",
"authors": "Yonathan Wijaya Gustian, Ben Rahman, Djarot Hindarto, Alessandro Benito Putra Bayu Wedha",
"venue": "SinkrOn",
"cites": 12,
"link": "https://doi.org/10.33395/sinkron.v8i2.12394",
"tasks": "segmentation; detection; severity",
"models": "YOLO",
"datasets": "Own/insurer data",
"reported_metrics": "precision 70%",
"techniques": "part localisation + damage",
"seed": false,
"abstract": "This journal presents a technique for detecting scratches, cracks and other damage to car bodies using machine learning methods. This method is used to improve process efficiency and checking accuracy and can also reduce the cost and time required for manual inspection. The method includes collecting image datasets of cars in good and damaged condition, followed by preprocessing and segmentation to separate damaged or damaged car parts. not broken. Then, it is followed by a deep learning algorithm, namely You Only Look Once, or Faster Region-based Convolutional Neural Networks, which is used to build a detection model. The model is trained and tuned using the collected data, then evaluated using the test data to measure the accuracy and precision of the detection results. The experimental results show that the proposed method achieves high accuracy and efficiency in detecting scratches, cracks, and other defects on the car body, with precision of an average of more than 70%. This method provides a promising approach to improving the car body inspection process which can be used by taxi companies to help inspect and maintain vehicles more quickly and accurately, to help with insurance, avoid accidents and so on."
},
{
"n": 14,
"year": 2023,
"title": "Deep Learning Based Car Damage Classification and Cost Estimation",
"authors": "Namam A. Mohammed, Moayad Yousif Potrus, Abbas M. Ali",
"venue": "ZANCO Journal of Pure and Applied Sciences",
"cites": 11,
"link": "https://doi.org/10.21271/zjpas.35.1.1",
"tasks": "detection; classification; severity",
"models": "Mask R-CNN",
"datasets": "Own/insurer data",
"reported_metrics": "accuracy 98.5%",
"techniques": "instance segmentation masks; severity / cost estimation",
"seed": false,
"abstract": "Due to the increasing number of people driving cars, the number of insurance claims has also increased. This process involves the manual assessment of the vehicle by an insurance company's service engineer, as well as the physical inspection by a licensed insurance company representative. An end-to-end solution has been proposed that would allow the customer and the insurance company to automate the process of recognizing the damaged area in the vehicles and estimating the cost of the damage. It would allow them to get a better understanding of the condition of the vehicle. For this purpose, A deep learning, Mask Region-based Convolutional Neural Network (Mask RCNN) model was utilized in this work to classify vehicle damages costs. Two Mask RCNN models were utilized, the first one was used to detect the sides of the vehicle, which will affect damage cost estimation and the second was used to find the area of the damage. The Experimental work shows that the proposed model gives reasonable results to estimate the cost of the damage. We achieve an accuracy of 98.5% with the combination of the two Mask RCNN models. And showed that Mask RCNN has a promising result to detect the area of the damage in the car."
},
{
"n": 15,
"year": 2025,
"title": "CAR DAMAGE DETECTION USING DEEP LEARNING",
"authors": "Indukuri, Rahul Varma, Sr.",
"venue": "CSUSB ScholarWorks (California State University, San Bernardino)",
"cites": 0,
"link": "https://openalex.org/W7112198689",
"tasks": "detection; classification",
"models": "SSD; Inception; MobileNet; Transfer learning",
"datasets": "COCO-car-damage",
"reported_metrics": "accuracy 94%",
"techniques": "data augmentation; transfer learning / pre-training; lightweight / mobile deployment; grounding / text-guided; damaged-vs-intact pre-filter",
"seed": false,
"abstract": "Growing vehicle usage has resulted in a notable increase in road accidents, so it is imperative to have effective systems for identifying and evaluating vehicle damage. This work aims to create a computer vision and deep learning-based automated car damage detection system. This project's main goal is to develop a model that, using visual cues, can categorize car photos as either damaged or undamaged. The algorithm operates in two steps: first, determining whether the picture features an automobile; then, it classifies the state of the car—damaged or undamaged. We thus employ the InceptionV3 model for damage classification and the MobileNet SSD (Single Shot Multibox Detector) for vehicle detection to do this. The image contains an automobile detected using a pre-trained object detection model called MobileNet SSD. Downloaded automatically during the procedure, the pre-trained set of weights injected into the model is once a car is spotted. The image is then fed to the InceptionV3 model, which has been optimized specifically for damage detection. Using a dataset of automotive photos, where every image is tagged as either damaged or undamaged, the InceptionV3 model was trained. With a 94% accuracy, the algorithm forecasts the car's damage status, proving great real-time predictive efficiency. Training uses a dataset of manually labeled car photos for damage and undamaged categories. The model detects vehicles using the COCO dataset labels; the car is classified as class 7 there"
},
{
"n": 16,
"year": 2026,
"title": "CropResMoE-50: A Region-Aware Mixture-of-Experts Framework for Fine-Grained Vehicle Damage Detection and Semi-Automated Annotation",
"authors": "Kam Ching Cheng, Yong Poh Yu, Tong Ming Lim",
"venue": "Journal of Artificial Intelligence and Technology",
"cites": 0,
"link": "https://doi.org/10.37965/jait.2026.0920",
"tasks": "detection; classification",
"models": "ResNet",
"datasets": "CarDD; COCO-car-damage",
"reported_metrics": "precision 87.7%",
"techniques": "data augmentation; active / semi-supervised learning; explainability (Grad-CAM)",
"seed": false,
"abstract": "Accurate detection of car damages such as dents, scratches, and cracks is vital for improving the efficiency and consistency of vehicle insurance claim processes. Traditional assessment methods rely heavily on manual inspections, which are labour-intensive and prone to subjectivity. This paper introduces a three-tiered progression of Mixture-of-Experts (MoE)-based classifiers trained on the CarDD dataset, which comprises 4,000 annotated images in COCO format. Starting with a baseline RawMoE model using flattened image vectors, we subsequently enhance performance through ResMoE-50, which integrates ResNet-50 for deep feature extraction. Finally, we propose CropResMoE-50, a region-aware hybrid architecture that applies object-level cropping and scale-specific analysis to improve spatial localization and classification precision. CropResMoE-50 achieves 89.30% test accuracy and attains average precision scores of 87.70%, 93.45%, and 98.12% for small, medium, and large objects, respectively. To extend model utility, we introduce a semi-automated labeling pipeline that integrates ChromaDB for retrieval-augmented learning. This hybrid mechanism enhances interpretability and supports pseudo-labeling under uncertainty by blending MoE predictions with memory-based exemplar retrieval. Experimental results demonstrate that the system can annotate new damage crops with minimal human oversight, making it suitable for real-world insurance triage and dataset expansion. The overall design pre"
},
{
"n": 17,
"year": 2023,
"title": "Assessment of Intelligent Vehicle Damage and Cost Estimator for Insurance Companies",
"authors": "S. Ramesh Babu B. Naga Rajesh",
"venue": "Journal of Propulsion Technology",
"cites": 0,
"link": "https://doi.org/10.52783/tjjpt.v44.i4.1003",
"tasks": "detection; severity",
"models": "VGG; Inception",
"datasets": "Own/insurer data",
"reported_metrics": "accuracy 94.56%",
"techniques": "severity / cost estimation",
"seed": false,
"abstract": "Now a day’s many accidents are taking place, Suppose if an accident had taken place and they want to claim with the respective car insurer they need to likely go through the process of having a person inspect the car's damage at your home or a repair centre and then needing to wait for an official estimate and approval to go ahead with getting the necessary repair work done. While this traditional process can be lengthy and keep your car off the road longer than you'd like, artificial intelligence can now play a role in speeding up the process by taking over the appraiser's work. Vehicle damage detection is used to reduce claims leakage during insurance processing. Visual inception and validation are usually done. As it takes a long time because a person needs to come and inspect the damage. Here we are trying to automate the procedure. Using this automation, we can avoid time conception for the insurance claim procedure. In this we are using the VGG16 model to predict the damage of vehicle. We get the accuracy of 94.56% of VGG16."
},
{
"n": 18,
"year": 2021,
"title": "Damaged Vehicle Parts Detection Platforms using Deep Learning Technique",
"authors": "Kundjanasith Thonglek, Norawit Urailertprasert, Patchara Pattiyathanee, Chantana Chantrapornchai",
"venue": "ECTI Transactions on Computer and Information Technology (ECTI-CIT)",
"cites": 9,
"link": "https://doi.org/10.37936/ecti-cit.2021153.223151",
"tasks": "detection; classification; severity",
"models": "Inception",
"datasets": "Own/insurer data",
"reported_metrics": "accuracy 93.28%; accuracy 98.47%",
"techniques": "severity / cost estimation; part localisation + damage",
"seed": false,
"abstract": "Automatic vehicle damage detection platform can increase the market value of car insurance. The es- timation process is usually manual and requires hu- man experts and their time to evaluate the damage cost. Intelligent Vehicle Accident Analysis (IVAA) system provides an artificial intelligence as a service (AIaaS) for building a system that can automatically assess vehicle parts’ damage and severity level. The insurance company can adopt our service to build the application to speedup the claiming process. There are four main elements in the service system which support four stakeholders in an insurance company: insurance experts, data scientists, operators and field employees. Insurance experts utilize the data label- ing tool to label damaged parts of a vehicle in a given image as a training data building process. Data scientists iterate to the deep learning model build- ing process for continuous model updates. Opera- tors monitor the visualization system for daily statis- tics related to the number of accidents based on lo- cations. Field employees use LINE Official integra- tion to take a photo of damaged vehicle at the acci- dent site and retrieve the repair estimation. IVAA is built on the docker image which can scale-in or scale- out the system depend on utilization efficiently. We deploy the Faster Region-based convolutional neural network, along with residual Inception network to lo- calize the damage region and classify into 5 damage levels for a vehicle part. The"
},
{
"n": 19,
"year": 2020,
"title": "Vehicle-Damage-Detection Segmentation Algorithm Based on Improved Mask RCNN",
"authors": "Qinghui Zhang, Xianing Chang, Shanfeng Bian",
"venue": "IEEE Access",
"cites": 152,
"link": "https://doi.org/10.1109/access.2020.2964055",
"tasks": "segmentation; detection",
"models": "Mask R-CNN; ResNet; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; instance segmentation masks; small-object / multi-scale features; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Traffic congestion due to vehicular accidents seriously affects normal travel, and accurate and effective mitigating measures and methods must be studied. To resolve traffic accident compensation problems quickly, a vehicle-damage-detection segmentation algorithm based on transfer learning and an improved mask regional convolutional neural network (Mask RCNN) is proposed in this paper. The experiment first collects car damage pictures for preprocessing and uses Labelme to make data set labels, which are divided into training sets and test sets. The residual network (ResNet) is optimized, and feature extraction is performed in combination with Feature Pyramid Network (FPN). Then, the proportion and threshold of the Anchor in the region proposal network (RPN) are adjusted. The spatial information of the feature map is preserved by bilinear interpolation in ROIAlign, and different weights are introduced in the loss function for different-scale targets. Finally, the results of self-made dedicated dataset training and testing show that the improved Mask RCNN has better Average Precision (AP) value, detection accuracy and masking accuracy, and improves the efficiency of solving traffic accident compensation problems."
},
{
"n": 20,
"year": 2025,
"title": "Robust anomaly detection through multi-modal autoencoder fusion for small vehicle damage detection",
"authors": "Sara Khan, Mehmed Yüksel, Frank Kirchner",
"venue": "Machine Learning with Applications",
"cites": 1,
"link": "https://doi.org/10.1016/j.mlwa.2025.100794",
"tasks": "detection",
"models": "Ensemble",
"datasets": "",
"reported_metrics": "AUC 92%",
"techniques": "ensemble / model fusion; lightweight / mobile deployment",
"seed": false,
"abstract": "Wear and tear detection in fleet and shared vehicle systems is a critical challenge, particularly in rental and car-sharing services, where minor damage, such as dents, scratches, and underbody impacts, often goes unnoticed or is detected too late. Currently, manual inspection methods are the default approach, but are labour-intensive and prone to human error. In contrast, state-of-the-art image-based methods are less reliable when the vehicle is moving, and they cannot effectively capture underbody damage due to limited visual access and spatial coverage. This work introduces a novel multi-modal architecture based on anomaly detection to address these issues. Sensors such as Inertial Measurement Units (IMUs) and microphones are integrated into a compact device mounted on the vehicle’s windshield. This approach supports real-time damage detection while avoiding the need for highly resource-intensive sensors. We developed multiple variants of multi-modal autoencoder-based architectures and evaluated them against unimodal and state-of-the-art methods. Our multi-modal ensemble model with pooling achieved the highest performance, with a Receiver Operating Characteristic-Area Under Curve (ROC-AUC) of 92%, demonstrating its effectiveness in real-world applications. This approach can also be extended to other applications, such as improving automotive safety. It can integrate with airbag systems for efficient deployment and help autonomous vehicles by complementing other sensors in "
},
{
"n": 21,
"year": 2025,
"title": "Car Damage Detection Analysis using Deep Learning, Computer Vision Techniques",
"authors": "Neeta Verma, P. Santhosh Kumar, Ruchika Yadav, Vidisha Kumar",
"venue": "International Journal of Computer Applications",
"cites": 0,
"link": "https://doi.org/10.5120/icaidsc202404",
"tasks": "detection; classification",
"models": "Transfer learning",
"datasets": "",
"reported_metrics": "accuracy 94%",
"techniques": "transfer learning / pre-training",
"seed": false,
"abstract": "Car damage detection is an important field of research due to its potential application in accident prevention, insurance claims, and law enforcement.In this paper, we propose a deep learning approach for car damage detection using image analysis.We use a pre-trained convolutional neural network (CNN) model to extract features from the input images and then train a support vector machine (SVM) classifier on the extracted features.The proposed approach is evaluated on a publicly available dataset of car damage images and achieves a high accuracy of 94% in car damage detection.Our results demonstrate the effectiveness of deep learning in car damage detection and have potential for real-world applications.."
},
{
"n": 22,
"year": 2021,
"title": "Automatic Car Damage detection by Hybrid Deep Learning Multi Label Classification",
"authors": "P. Ebby Darney",
"venue": "Journal of Artificial Intelligence and Capsule Networks",
"cites": 0,
"link": "https://doi.org/10.36548/jaicn.2021.4.006",
"tasks": "detection; classification",
"models": "ResNet; Transfer learning",
"datasets": "COCO-car-damage",
"reported_metrics": "accuracy 90.82%",
"techniques": "transfer learning / pre-training; lightweight / mobile deployment",
"seed": false,
"abstract": "Automating image-based automobile insurance claims processing is a significant opportunity. In this research work, car damage categorization that is aided by the hybrid convolutional neural network approach is addressed and hence the deep learning-based strategies are applied. Insurance firms may leverage this paper's design and implementation of an automobile damage classification/detection pipeline to streamline car insurance claim policy. Using deep convolutional networks to detect car damage is now possible because of recent improvements in the artificial intelligence sector, mainly due to less computation time and higher accuracy with a hybrid transformation deep learning algorithm. In this paper, multiclass classification proposed to categorize the car damage parts such as broken headlight/taillight, glass fragments, damaged bonnet etc. are compiled into the proposed dataset. This model has been pre-trained on a wide-ranging and benchmark dataset due to the dataset's limited size to minimize overfitting and to understand more common properties of the dataset. To increase the overall proposed model’s performance, the CNN feature extraction model is trained with Resnet architecture with the coco car damage detection datasets and reaches a higher accuracy of 90.82%, which is much better than the previous findings on the comparable test sets."
},
{
"n": 23,
"year": 2026,
"title": "Lightweight Vehicle Damage Detection With NMS‐Free YOLOv10s and AKConv",
"authors": "Haichang Jiang, Taozhu Zhou, Yingqian Lao, Chenxin Nie",
"venue": "IET Image Processing",
"cites": 0,
"link": "https://doi.org/10.1049/ipr2.70402",
"tasks": "detection",
"models": "YOLOv10; Xception",
"datasets": "",
"reported_metrics": "mAP 81.48%",
"techniques": "lightweight / mobile deployment",
"seed": false,
"abstract": "ABSTRACT This study addresses the inefficiency and human error susceptibility of traditional vehicle damage assessment methods by proposing an intelligent detection model based on an enhanced YOLOv10s algorithm. The model incorporates an adaptive Kernel convolution (AKConv) module to replace standard convolutions, strengthening its capability to characterise multi‐scale and irregular damage features. Leveraging YOLOv10s' end‐to‐end training mechanism that eliminates non‐maximum suppression requirements, computational complexity is reduced. Experimental validation on a 14‐category vehicle damage dataset shows the optimised YOLOv10s‐AKConv model achieves approximately 6% lower FLOPs and 2% fewer parameters while maintaining lightweight properties. The model attains 81.48% mAP50‐95 on the test set—a 1% improvement over the baseline‐ and demonstrates exceptional robustness in extreme conditions. These advancements significantly enhance damage recognition accuracy and efficiency, establishing a viable technical approach for practical applications such as insurance claim assessment."
},
{
"n": 24,
"year": 2026,
"title": "An intelligent approach for automated vehicle damage classification",
"authors": "Shimaa M. Ouf, Bassant Lotfy, Soha Ahmed",
"venue": "Scientific Reports",
"cites": 0,
"link": "https://doi.org/10.1038/s41598-026-61488-5",
"tasks": "detection; classification; severity",
"models": "YOLOv8; YOLOv5",
"datasets": "",
"reported_metrics": "precision 87%; precision 90.7%",
"techniques": "severity / cost estimation",
"seed": false,
"abstract": "Manual examination using imitative assessment methods requires considerable investment in time, effort, resources, and funds. Furthermore, these assessments often exhibit inconsistencies. Therefore, this study introduces a framework for accurately detecting and classifying vehicle damage and estimating repair costs using deep learning and machine learning approaches. In this study, YOLOv5 and YOLOv8 models were applied to detect and classify vehicle damage types, such as broken lamps, glass shatters, cracks, scratches, and dents. Repair costs were predicted using the XGBoost machine learning model. The dataset used for damage detection, classification, and cost prediction was created from scratch. It was developed by identifying the types of damage and generating bounding box coordinates around damaged areas. After extracting the bounding box coordinates and damage types, additional features were incorporated. The YOLOv8 model outperformed YOLOv5 in both detecting and classifying vehicle damage, achieving a precision of 87% on the validation dataset and mean absolute precision 90.7% on the test dataset. The XGBoost model achieved an R² score of 97.28% and a Mean Absolute Error (MAE) of 128.50. These results confirm that the proposed framework enhances the precision of vehicle damage detection and classification while accelerating assessment and ensuring its effectiveness in real-world scenarios."
},
{
"n": 25,
"year": 2026,
"title": "AI-Based Vehicle Damage and Safety Analysis",
"authors": "M. Karthika M. Vasuki",
"venue": "Zenodo (CERN European Organization for Nuclear Research)",
"cites": 0,
"link": "https://doi.org/10.5281/zenodo.21491792",
"tasks": "detection; classification; severity",
"models": "YOLOv10",
"datasets": "",
"reported_metrics": "mean average precision 88.7%",
"techniques": "severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "Vehicle damage assessment is a critical requirement in automobile maintenance, insurance claim verification, and fleet management operations. Conventional inspection methods rely on manual evaluation, which is time-consuming, inconsistent, and highly susceptible to human error. This paper proposes an intelligent framework that automates vehicle damage detection and severity estimation using Deep Learning and Computer Vision techniques. The system employs YOLOv10 as the core detection architecture to accurately identify and localize damaged regions within vehicle images. Four primary damage categories are recognized by the framework, namely scratches, dents, cracks, and broken vehicle components. Detected damage instances are further classified into three severity levels — minor, moderate, and severe — to support informed repair and claim decisions. Prior to detection, images undergo a structured preprocessing pipeline involving resizing, normalization, and noise filtering to ensure consistent input quality. The framework additionally supports automated report generation, enabling seamless integration into vehicle service platforms and insurance processing workflows. Experimental evaluation demonstrates that the proposed system achieves a mean average precision of 88.7% with an inference time of 15 milliseconds per image. The results confirm that the proposed approach offers a scalable, reliable, and efficient solution for AI-driven automated vehicle damage analysis."
},
{
"n": 26,
"year": 2025,
"title": "A System for Automated Vehicle Damage Localization and Severity Estimation Using Deep Learning",
"authors": "Geetha M, Chaitra KA",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 0,
"link": "https://doi.org/10.55041/ijsrem52093",
"tasks": "detection; classification; severity",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "accuracy 91%",
"techniques": "severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "ABSTRACT This paper introduces a sophisticated and comprehensive system for automated vehicle damage detection and severity estimation by leveraging the advanced YoloV8 deep learning model. The proposed solution operates through a web-based interface developed using Python, HTML, CSS, and JavaScript, hosted on the Flask framework. It supports three operational modes: image-based, video-based, and webcam-based predictions, thereby enabling real-time as well as post-incident evaluations. The model is trained on a custom-labeled dataset of 778 images that cover eight distinct vehicle damage categories. Achieving an accuracy rate of 91%, the system enhances reliability in assessing vehicle damages. Designed for scalability and versatility, this system offers valuable applications in domains such as automotive insurance claims, vehicle repair cost assessments, accident investigation, and fleet management. Keywords: Vehicle damage detection, YoloV8, deep learning, severity estimation, automated assessment, image classification, real-time analysis, web-based interface, Flask framework."
},
{
"n": 27,
"year": 2025,
"title": "CAR Damage Price Predictor",
"authors": "G.Uma Maheswari, Jabez Jerin J, John Rathinam, P Sunil",
"venue": "International Research Journal on Advanced Engineering Hub (IRJAEH)",
"cites": 0,
"link": "https://doi.org/10.47392/irjaeh.2025.0073",
"tasks": "segmentation; detection; classification; severity",
"models": "Mask R-CNN; Faster R-CNN; Transfer learning",
"datasets": "",
"reported_metrics": "accuracy 95%",
"techniques": "transfer learning / pre-training; instance segmentation masks; severity / cost estimation; lightweight / mobile deployment; reflection / false-positive handling",
"seed": false,
"abstract": "The automotive repair industry is evolving, and with that comes increasing demand for damage assessment's accuracy and efficiency. In this project, we propose a web platform for predicting car damage severity and repair costs using state-of-the-art machine learning and deep learning techniques. The platform uses Mobile Net-a light-weight convolutional neural network-for efficient and accurate image classification. The website allows users to upload uploaded images of damaged cars to view fast evaluation on damages classified into either high, medium, or low, along with detailed estimates of repair costs. The system allows a smooth upload with SQLite for safe data management while providing better prediction using transfer learning and pre-trained models. Faster R-CNN and Mask R-CNN are also applied for precise localization and instance segmentation. This novel method is envisioned as a technology that will transform car repair by providing a credible, effective, and accessible tool for automated damage assessment that lets vehicle owners decide with time and resource savings. The platform achieved remarkable diagnostic accuracy at up to 95%, thus significantly reducing false positives and negatives while offering advice to the car owner and car repair professionals."
},
{
"n": 28,
"year": 2026,
"title": "Cross-Architecture Performance Evaluation of Transfer Learning Models for Multi-Class Vehicle Damage Severity Classification",
"authors": "Mochammad Fatih Ulumuddin, Anggay Luri Pramana",
"venue": "SinkrOn",
"cites": 0,
"link": "https://doi.org/10.33395/sinkron.v10i2.15939",
"tasks": "classification; severity",
"models": "ResNet; VGG; MobileNet; Transfer learning",
"datasets": "",
"reported_metrics": "accuracy 75.76%",
"techniques": "transfer learning / pre-training; severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "Automated vehicle damage classification supports objectivity and scalability in insurance claim processing and digital inspection systems; however, prior studies often report performance improvements without controlled experimental settings or statistical validation, limiting methodological reliability. This study establishes a statistically controlled cross-architecture evaluation framework to determine whether pretrained convolutional neural networks significantly outperform a custom baseline model in multi-class vehicle damage classification. A dataset of 891 labeled vehicle images categorized into heavy, medium, light, and normal damage was partitioned using stratified sampling (70% training, 15% validation, 15% testing). Four architectures Baseline (CustomCNN), VGG16, ResNet50, and MobileNetV2 were trained under identical preprocessing and optimization settings with two training durations (30 and 50 epochs). Five-fold cross-validation and paired t-test analysis were applied to assess statistical significance. At 30 epochs, MobileNetV2 achieved the highest accuracy (75.76%), while at 50 epochs VGG16 obtained the best performance (78.03%). Extending training duration did not produce statistically significant improvement (p > 0.05). Pretrained architectures significantly outperformed the baseline model, whereas ResNet50 did not demonstrate superior performance. The novelty of this study lies in its statistically validated comparative framework. Although limited by moderate "
},
{
"n": 29,
"year": 2023,
"title": "Automatic damaged vehicle estimator using enhanced deep learning algorithm",
"authors": "Jihad Qaddour, Syeda Ayesha Siddiqa",
"venue": "Intelligent Systems with Applications",
"cites": 31,
"link": "https://doi.org/10.1016/j.iswa.2023.200192",
"tasks": "segmentation; detection; classification; severity",
"models": "Mask R-CNN; ResNet; VGG; Inception; Transfer learning",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; instance segmentation masks; severity / cost estimation; lightweight / mobile deployment; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Claim leakage costs insurance companies millions of dollars each year because of the disparity between the cost spent by allowance businesses and the accurate quantity that must be reimbursed. As a result, processing claims for identifying and classifying automobile damages takes time and is costly for insurance providers. In this paper, we used an improved Mask R-CNN method, which has a significant research benefit of object detection, to automatically detect, identify, and categorize car damage sites in traffic incidents. To detect and label an image of a damaged vehicle, we used a combination of deep learning, transfer learning, Mask R-CNN, and instance segmentation. In addition, a web-based automatic claim estimator can accept photographs from the user and determine the position and degree of the damage automatically. Furthermore, three different pre-trained models, namely inception ResNetV2, VGG-16, and VGG-19, were used to aid quick convergence. Finally, comparative performance assessments employ several evaluation measures such as precision, recall, F1 score, accuracy, loss function, and confusion matrices based on the three pre-trained models. The empirical results reveal that the proposed method not only recognizes damaged vehicles but also locates them and determines their severity level which accomplishes the study's objective of automatically locating and classifying car damage. According to the data, employing Mask-RCNN with pre-trained Inception ResNetV2 outperf"
},
{
"n": 30,
"year": 2024,
"title": "Automated Car Damage Assessment Using Computer Vision: Insurance Company Use Case",
"authors": "Sergio A. Pérez-Zarate, Daniel Corzo-García, José Luis Pro-Martín, Juan A. Álvarez-García",
"venue": "Applied Sciences",
"cites": 17,
"link": "https://doi.org/10.3390/app14209560",
"tasks": "detection",
"models": "YOLOv8; YOLOv5; Ensemble",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "ensemble / model fusion",
"seed": false,
"abstract": "Automated car damage detection using computer vision techniques has been studied using several datasets, but real cases for insurance companies are usually dependent on private methods and datasets. Furthermore, there are no metrics or standardized processes that describe the situation in which the company analyzes the customer’s images, the models used for the inference, and the results. We perform extensive experiments to show that our proposal, an ensemble of 10 deep learning detectors based on YOLOv5, improves the state-of-the-art not only in terms of typical metrics but also in terms of inference speed, allowing scalability to thousands of instances per minute. A comparison with YOLOv8 is carried out, showing the differences between both ensembles. Furthermore, a dataset called TartesiaDS, labeled under the supervision of professional appraisers from insurance companies, is available to the community for evaluation of future proposals."
},
{
"n": 31,
"year": 2024,
"title": "GroundingCarDD: Text-Guided Multimodal Phrase Grounding for Car Damage Detection",
"authors": "Md Jahid Hasan, Agustinus Nalwan, Kok-Leong Ong, Hamed Jahani",
"venue": "IEEE Access",
"cites": 12,
"link": "https://doi.org/10.1109/access.2024.3506563",
"tasks": "detection; classification",
"models": "YOLOv9; DETR",
"datasets": "CarDD",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); grounding / text-guided; reflection / false-positive handling",
"seed": false,
"abstract": "The rising incidence of vehicle accidents and the growth of online car marketplaces underscore the critical need for efficient and accurate car damage inspection in rental services, insurance claims processing, and vehicle maintenance. Traditional manual inspection methods are time-consuming, prone to errors, and vulnerable to fraud, emphasizing the necessity for automated solutions. While existing automated object detection models are effective in many scenarios, they struggle to accurately identify minor damages on car bodies and distinguish actual damage from car features, reflections, or shadows that resemble damage. This limitation stems from a lack of effective methods to extract distinct features of minor damages and a lack of contextual understanding beyond visual cues. To address this challenge, we introduce GroundingCarDD, a text-guided multimodal phrase grounding framework that leverages both visual and textual information to precisely localize and classify car damages. By employing a context-aware attention mechanism to fuse visual and textual features, GroundingCarDD outperforms state-of-the-art methods. Specifically, it achieves a mean Average Precision (mAP) of 64.1 and makes only a miss detection rate of 14.4% on a combined dataset of public and curated private data, surpassing other state-of-the-art methods YOLOv9 and DETR. Furthermore, on the public CarDD dataset, it attains an AP50 of 80 and a recall of 86.7, exceeding the performance of existing models. Gr"
},
{
"n": 32,
"year": 2025,
"title": "On the Application of DiffusionDet to Automatic Car Damage Detection and Classification via High-Performance Computing",
"authors": "Vito Arconzo, Gerardo Gorga, Gonzalo Gutierrez, Ahmed Omar",
"venue": "Electronics",
"cites": 3,
"link": "https://doi.org/10.3390/electronics14071362",
"tasks": "detection; classification",
"models": "Swin",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "synthetic data / GAN; transformer backbone",
"seed": false,
"abstract": "Claim management is a critical process for insurance companies, requiring fairness, transparency, and efficiency to maintain policyholder trust and minimize financial impact. In our previous work, we introduced Insoore AI, an insurtech solution leveraging deep learning-based computer vision to automate car damage recognition and localization from user-provided pictures. While this approach demonstrated the potential of AI in claims management, it faced limitations in terms of performance and computational efficiency due to resource constraints. In this study, we present an improved version of Insoore AI, enabled by the High-Performance Computing (HPC) resources offered by the Booster module of LEONARDO HPC system located at the CINECA datacenter in Bologna, Italy. By leveraging the advanced computational capabilities of the above-mentioned HPC infrastructure, we trained larger and more complex deep learning models, processed higher-resolution images, and significantly reduced training and inference times. Our results show marked performance improvements in terms of damage detection, paving the way for more efficient, more effective and scalable claims management solutions. This work underscores the transformative potential of HPC resources in advancing AI-driven innovations in the insurance sector and is to be regarded as an improvement on the contribution of our previous work, enabled by relying on the DiffusionDet architecture and on a Swin Transformer backbone to solve the"
},
{
"n": 33,
"year": 2024,
"title": "Implementation of Car Damage Detection in Insurance Claims",
"authors": "Prof. Soniya Dhotre",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 1,
"link": "https://doi.org/10.55041/ijsrem34585",
"tasks": "detection; severity",
"models": "Mask R-CNN",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "instance segmentation masks",
"seed": false,
"abstract": "The automotive industry is witnessing a growing demand for advanced safety and maintenance systems to enhance vehicle safety and reduce maintenance costs. Car damage detection plays a pivotal role in achieving these objectives. This abstract provides an overview of the concept of car damage detection, its importance, and the technologies and methodologies involved. Car damage detection refers to the process of identifying and assessing physical damage to a vehicle, such as dents, scratches, cracks, or structural deformities. It is essential for various stakeholders, including car owners, insurance companies, and automotive service providers, as it enables timely repair and maintenance, facilitates insurance claims, and ensures road safety. Key Words: CNN, Mask R-CNN, Image Analysis, Computer Vision, Machine Learning, Deep-Learning."
},
{
"n": 34,
"year": 2025,
"title": "Automated Vehicle Damage Detection and Cost Estimator for Insurance Companies",
"authors": "Chinthaprthi Chethan",
"venue": "International Journal for Research in Applied Science and Engineering Technology",
"cites": 1,
"link": "https://doi.org/10.22214/ijraset.2025.67989",
"tasks": "detection; severity",
"models": "YOLOv5",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation; small-object / multi-scale features; lightweight / mobile deployment",
"seed": false,
"abstract": "Using state-of-the-art deep learning algorithms, the \"Automated Vehicle Damage Detection & Cost Estimator for Insurance Companies\" project seeks to transform the vehicle insurance industry by automating the damage assessment process. This AI-powered system uses YOLOv5, one of the most sophisticated object identification models available, to effectively analyse and evaluate vehicle damage, including that of two-wheelers and four-wheelers. The model can precisely identify, categorize, and assess various forms of damage, such as scratches, dents, cracks, and shattered pieces, because it has been thoroughly trained on a vast dataset of car damages. This high degree of accuracy lowers the possibility of errors in computations and false claims by guaranteeing that even small damages are accurately identified. Insurance firms may improve the speed of their claims processing, reduce human error, and produce accurate repair cost estimates based on real-time damage analysis by putting deep learning and computer vision technology into practice. By automating the evaluation process, the system will do away with the necessity for manual inspections, which are frequently laborious, arbitrary, and prone to errors. Additionally, the system would forecast repair costs based on variables like labor prices, spare parts, and vehicle depreciation by integrating repair cost estimating models. This will enable insurance companies to give equitable and transparent claim settlements. The entire insur"
},
{
"n": 35,
"year": 2025,
"title": "DamageNet: A Dilated Convolution Feature Pyramid Network Mask R‑CNN for Automated Car Damage Detection and Segmentation",
"authors": "Nazbek Katayev, Zhanna Yessengaliyeva, Zhazira Kozhamkulova, Zhanel Bakirova",
"venue": "International Journal of Advanced Computer Science and Applications",
"cites": 1,
"link": "https://doi.org/10.14569/ijacsa.2025.0160521",
"tasks": "segmentation; detection; classification",
"models": "",
"datasets": "VehiDE",
"reported_metrics": "mean Average Precision 85.7%; mIoU 82.3%",
"techniques": "instance segmentation masks; small-object / multi-scale features; lightweight / mobile deployment",
"seed": false,
"abstract": "Automated and precise assessment of vehicle damage is critical for modern insurance processing, accident analysis, and autonomous maintenance systems. In this work, we introduce DamageNet, a unified deep instance segmentation framework that embeds a multi‑rate dilated‑convolution context module within a Feature Pyramid Network (FPN) backbone and couples it with a Region Proposal Network (RPN), RoI‑Align, and parallel heads for classification, bounding‑box regression, and pixel‑level mask prediction. Evaluated on the large‑scale VehiDE dataset comprising 5 200 high‑resolution images annotated for dents, scratches, and broken glass, DamageNet achieves a mean Average Precision (mAP) of 85.7% for damage localization and a mean Intersection over Union (mIoU) of 82.3% for segmentation, outperforming baseline Mask R‑CNN by 6.2 and 7.8 percentage points, respectively. Ablation studies confirm that the dilated‑convolution module, multi‑scale fusion in the FPN, and post‑processing refinements each contribute substantially to segmentation fidelity. Qualitative results demonstrate robust delineation of both subtle scratch lines and extensive panel deformations under diverse lighting and occlusion conditions. Although the integration of atrous convolutions introduces a modest inference overhead, DamageNet offers a significant advancement in end‑to‑end vehicle damage analysis. Future extensions will investigate lightweight dilation approximations, dynamic rate selection, and semi‑supervise"
},
{
"n": 36,
"year": 2023,
"title": "Deep Learning-based Damage Detection and Repair Cost Estimation for Automobiles",
"authors": "Narra Suma, N. Savitha",
"venue": "Zenodo (CERN European Organization for Nuclear Research)",
"cites": 1,
"link": "https://doi.org/10.5281/zenodo.8081873",
"tasks": "detection; severity",
"models": "MobileNet; Transfer learning",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "In the automotive sector, estimating the cost to repair damaged vehicles is a critical duty. In this project, we present a technique for estimating maintenance costs that makes use of the cutting-edge deep learning architecture MobileNetV2. With the use of a sizable dataset of photos of damaged vehicles and related repair costs, we fine-tune the pre-trained MobileNetV2 model. For insurance companies, repair facilities, and automobile manufacturers, the suggested method can be used as a cost-effective and practical option to determine the expense of repairing damaged vehicles."
},
{
"n": 37,
"year": 2024,
"title": "Analysis Performance of One-Stage and Two Stage Object Detection Method for Car Damage Detection",
"authors": "Harum Ananda Setyawan, Alhadi Bustamam, Rinaldi Anwar Buyung",
"venue": "International Journal of Advanced Computer Science and Applications",
"cites": 1,
"link": "https://doi.org/10.14569/ijacsa.2024.01507103",
"tasks": "detection",
"models": "YOLOv8; YOLOv5; ResNet; MobileNet",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "instance segmentation masks; lightweight / mobile deployment",
"seed": false,
"abstract": "The large use of private cars is directly proportional to the number of insurance claims. Therefore, insurance companies need a breakthrough or new approach that is more effective and efficient to be able to compete for the trust of their customers. One approach that can be taken is to use artificial intelligence to detect damage to the car body to speed up the claims process. In this research, several experiments will be carried out using various types of models, namely Mask-R-CNN, ResNet50, MobileNetv2, YOLO-v5, and YOLO-v8 to detect damage to the car body. Furthermore, in the experiments that were carried out, the best results were obtained using the YOLO-v8x model with precision, recall, and F1-score values of 0.963, 0.951, and 0.936 respectively."
},
{
"n": 38,
"year": 2026,
"title": "Predicting Vehicle Damage Localization and Severity Estimation using Deep learning.",
"authors": "G. Jagan Naik, A. Hemalatha, Dr. A. Vijender",
"venue": "JOURNAL OF ADVANCE AND FUTURE RESEARCH",
"cites": 0,
"link": "https://doi.org/10.56975/jaafr.v4i6.512123",
"tasks": "detection; severity",
"models": "YOLOv8",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation; part localisation + damage; lightweight / mobile deployment",
"seed": false,
"abstract": "The manual inspections currently performed can be time consuming, inconsistent and, as with most human activities, present a huge risk of error. that are currently done can be extremely time-consuming, do not have any consistency and as with most human-driven activities, a huge possibility of error. automated system to detect and label vehicle damage severity using a YOLOv8 deep learning model. tackle such limitations, using a YOLOv8 deep learning model. It locates the damaged part, recognizes different types of vehicle damage and gives an estimate of the severity level of that damage based on input images, videos or live webcam feeds. different types of vehicle damages and gives an estimated level of severity of those damages based on input images, videos or live webcam streams. The model was trained using 778 labeled images of 8 damage types, the average detection rate is around 91%. using778 labeled images of 8 kinds of damages, the average detection rate is approximately 91%. The proposed system reduces the human effort involved, increases the consistency of assessment, and provides a solution for insurance companies and automobile service stations to perform rapid and accurate damage assessment. This proposed system reduces the human effort involved, increases consistency in assessment and provides a solution for insurance companies and automobile service stations to perform swift and accurate assessment of damages."
},
{
"n": 39,
"year": 2026,
"title": "AI-Based Vehicle Damage Detection, Cost Estimation, and Insurance Claim Prediction System",
"authors": "Sangeetha Navale, Prachi Raut, Prachi Ukey, Atharva Kulkarni",
"venue": "International Journal of Electrical Electronics and Computer Systems",
"cites": 0,
"link": "https://doi.org/10.65521/ijeecs.v15i1s.3100",
"tasks": "detection; classification; severity",
"models": "YOLOv8; Transfer learning",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; severity / cost estimation",
"seed": false,
"abstract": "Road accidents result in billions of rupees worth of vehicle damage every year, yet the process of assessing that damage and processing insurance claims has remained stubbornly manual, slow, and error-prone. This paper presents the design, implementation, and evaluation of an AI-based vehicle damage detection and cost estimation system that automates the entire assessment pipeline through the integration of deep learning and machine learning. The proposed system is a Flask-based web application in which a pre-trained YOLOv8 object detection model identifies damaged vehicle components from user-uploaded photographs, including parts such as bonnets, bumpers, doors, and fenders. Detected parts are cross-referenced against a structured JSON pricing database to compute a repair cost estimate using a part-wise summation formula. A rule-based module predicts probable internal damages from observed external patterns, and a trained scikit-learn classification model determines insurance eligibility and computes coverage. The system classifies overall damage severity into minor, moderate, and major categories and produces smart repair and financial recommendations. Experimental results demonstrate that the YOLOv8 model achieves a mean Average Precision (mAP@50) of 0.79 across seven damage classes, with a cost estimation mean absolute percentage error of 11.4% relative to authorised workshop quotations. The integrated pipeline reduces a conventionally multi-day assessment process to unde"
},
{
"n": 40,
"year": 2026,
"title": "DEEP LEARNING BASED VEHICLE DAMAGE ASSESSMENT SYSTEM",
"authors": "1Mrs. M. Anusha, 2K. Sravanthi, 3N. Sahithi, 4K. V. S. Janani",
"venue": "International Journal of Data Science and IoT Management System",
"cites": 0,
"link": "https://ijdim.com/journal/index.php/ijdim/article/download/1244/1182",
"tasks": "detection; classification; severity",
"models": "YOLO; Mask R-CNN",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "instance segmentation masks; severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "A Deep learning-based vehicle damage assessment system is an intelligent solution that automates the process of detecting and evaluating vehicle damage from images. The system uses advanced deep learning techniques, particularly convoluational neural Networks (CNNs) and object detection models such as YOLO or Mask R-CNN, to identify damaged areas including dents, scratches, cracks, and broken parts. After analyzing the uploaded vehicle images, the system classifies the type and severity of damage and generates an assessment report. This reduce the need for manual inspection, minimize human error, and speeds yup the insurance claim and repair estimation process. The proposed system provides accurate, reliable, and cost-effectiv e damage assessment, making it beneficial for insurance companies, vehicle owners, and automobiles service centers. By leveraging artificial intelligence and computer vivion, system improves efficiency cocsistent evaluations, and enhances customer satisfaction in the vehicle inspection process. KEYWORDS: Deep learning, Vehicle Damage Assessment, computer vivion, convolutional Neural Networks (CNN), Object Detection, Damage Classification, insurance Claim Automation, Artificial Intelligence (AI)."
},
{
"n": 41,
"year": 2026,
"title": "DEVELOPING A MASK R-CNN MODEL FOR DETECTING CAR DAMAGE",
"authors": "Mr. MD. IMRAN,T. KEDAR,U. PARDHIV SAI,T. SUPRIYA,P. MANEESH",
"venue": "International Journal of Engineering Research and Science & Technology",
"cites": 0,
"link": "https://ijerst.org/index.php/ijerst/article/download/3045/2760",
"tasks": "segmentation; detection",
"models": "Mask R-CNN; Faster R-CNN",
"datasets": "COCO-car-damage",
"reported_metrics": "",
"techniques": "instance segmentation masks; lightweight / mobile deployment",
"seed": false,
"abstract": "general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing us to estimate human poses in the same framework. We show top results in all three tracks of the COCO suite of challenges, including instance segmentation, boundingbox object detection, and person keypoint detection. Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners. We hope our simple and effective approach will serve as a solid baseline and help ease future research in instance-level recognition."
},
{
"n": 42,
"year": 2020,
"title": "Assessing Car Damage using Mask R-CNN",
"authors": "Sarath P, Soorya M, Shaik Abdul Rahman A, S Suresh Kumar",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2004.14173v3",
"tasks": "detection; classification",
"models": "Mask R-CNN",
"datasets": "",
"reported_metrics": "precision 89.5%",
"techniques": "instance segmentation masks",
"seed": false,
"abstract": "Picture based vehicle protection handling is a significant region with enormous degree for mechanization. In this paper we consider the issue of vehicle harm characterization, where a portion of the classifications can be fine-granular. We investigate profound learning based procedures for this reason. At first, we attempt legitimately preparing a CNN. In any case, because of little arrangement of marked information, it doesn't function admirably. At that point, we investigate the impact of space explicit pre-preparing followed by tweaking. At last, we explore different avenues regarding move learning and outfit learning. Trial results show that move learning works superior to space explicit tweaking. We accomplish precision of 89.5% with blend of move and gathering learning."
},
{
"n": 43,
"year": 2024,
"title": "Automated vehicle damage classification using the three-quarter view car damage dataset and deep learning approaches",
"authors": "Donggeun Lee, Juyeob Lee, Eunil Park",
"venue": "Heliyon",
"cites": 22,
"link": "https://doi.org/10.1016/j.heliyon.2024.e34016",
"tasks": "classification",
"models": "ViT; ResNet; EfficientNet; MobileNet; DenseNet; Ensemble; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; ensemble / model fusion; lightweight / mobile deployment; damaged-vs-intact pre-filter; transformer backbone; dataset contribution",
"seed": false,
"abstract": "Automated procedures for classifying vehicle damage are critical in industries requiring extensive vehicle management. Despite substantial research demands, challenges in the field of vehicle damage classification persist due to the scarcity of public datasets and the complexity of constructing datasets. In response to these challenges, we introduce a Three-Quarter View Car Damage Dataset (TQVCD dataset), emphasizing simplicity in labeling, data accessibility, and rich information inherent in three-quarter views. The TQVCD dataset distinguishes class by vehicle orientation (front or rear) and type of damage while maintaining a three-quarter view. We evaluate performance using five prevalent pre-trained deep learning architectures-ResNet-50, DenseNet-160, EfficientNet-B0, MobileNet-V2, and ViT-employing a suite of binary classification models. To enhance classification robustness, we implement a model ensemble method to effectively mitigate individual model dependencies' deviations. Additionally, we interview three experts from the used-car platform to validate the necessity of a vehicle damage classification model using the corresponding dataset from an industrial perspective. Empirical findings underscore the dataset's comprehensive coverage of vehicle perspectives, facilitating efficient data collection and damage classification while minimizing labor-intensive labeling efforts."
},
{
"n": 44,
"year": 2022,
"title": "Vehicle Damage Detection and Classification Using Image Processing",
"authors": "B Mallikarjuna, Arun Kumar K L",
"venue": "International Journal of Advanced Research in Science Communication and Technology",
"cites": 11,
"link": "https://doi.org/10.48175/ijarsct-5414",
"tasks": "detection; classification; severity",
"models": "YOLO",
"datasets": "COCO-car-damage",
"reported_metrics": "",
"techniques": "lightweight / mobile deployment",
"seed": false,
"abstract": "Vehicle have an impact on people’s daily safety, and because there are so many different types and sizes of materials, it can be challenging to distinguish and detect the conditions around the vehicle. In this project, we looked into the matter of car damage classification and detection, which insurance providers can utilize to quickly automates the handling of vehicle insurance disputes. Deep convolutional networks can be used to detect car damage and with recent developments in computer vision, which are largely attributable to the implementation of quick, scalable, and entire trainable CNN. We manually gathered and annotated pictures of numerous online sources that showed various kinds of car damage. By analyzing the deep learning-based YOLO (you only look once) series target detection method, a recognition approach that relies on YOLOS is provided to achieve timely and efficient identification of the damage in the vehicle. The COCO dataset's base weights are used to train the model. 35-90 epochs are used to process the photos. The region of damage is highlighted in the final image using a color splash technique after processing. The approach would increase customer satisfaction while assisting in lowering the cost of processing insurance claims. Vendors of automobiles can do away with the labor-intensive manual damage assessment process. Additional vehicles will be priced accurately and transparently, along with any necessary repairs. It is also able to decrease misleadin"
},
{
"n": 45,
"year": 2021,
"title": "Deep Learning Based Car Damage Detection, Classification and Severity",
"authors": "",
"venue": "International Journal of Advanced Trends in Computer Science and Engineering",
"cites": 10,
"link": "https://doi.org/10.30534/ijatcse/2021/031052021",
"tasks": "detection; classification; severity",
"models": "YOLO; ResNet; VGG; DenseNet; Transfer learning",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; severity / cost estimation; damaged-vs-intact pre-filter",
"seed": false,
"abstract": "In the accident insurance industry, settling the claim is a time-consuming process since it is a manual process and there is a gap between the optimal and the actual settlement. Using deep learning models, we are not only trying to speed up the process but also provide better customer service and increase the profitability of insurance companies. In this paper we are using various pretrained models such as VGG 16, VGG 19, Resnet50 and Densenet and based on these models, selecting the best performing models. We initially check whether the car is damaged or not using the Resnet50 model and if it’s a damaged one we use the WPOD-net model to detect the license plate. To identify the damaged region, we use the YOLO model. At last, comes the damage severity which is implemented using the Densenet model. After implementing various models, we find out that transfer learning gives better results than fine-tuning. In addition to that we propose a framework that integrates all of this into one application and in turn helps in the automation of the insurance industry"
},
{
"n": 46,
"year": 2022,
"title": "A novel data processing approach to detect fraudulent insurance claims for physical damage to cars",
"authors": "Ahmet YÜCEL",
"venue": "journal of new results in science",
"cites": 6,
"link": "https://doi.org/10.54187/jnrs.1129440",
"tasks": "detection; classification",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "accuracy 76.56%",
"techniques": "lightweight / mobile deployment",
"seed": false,
"abstract": "Some automobile insurance companies use computerized auto-detection systems to expedite claims payment decisions for insured vehicles. Claims suspected of fraud are evaluated using empirical data from previously investigated claims. The main objective of this manuscript is to demonstrate a novel data processing system and its potential for use in data classification. The data processing approach was used to develop a machine learning-based sentiment classification model to describe property damage fraud in vehicle accidents and the indicators of fraudulent claims. To this end, Singular Value Decomposition-based components and correlation-based composite variables were created. Machine learning models were then developed, with predictors and composite variables selected based on standard feature selection procedures. Five machine learning models were used: Boosted Trees, Classification and Regression Trees, Random Forests, Artificial Neural Networks, and Support Vector Machines. For all models, the models with composite variables achieved higher accuracy rates, and among these models, the artificial neural network was the model with the highest accuracy performance at 76.56%."
},
{
"n": 47,
"year": 2026,
"title": "From Synthetic Data to Real-World Efficiency: Automated Vehicle Damage Detection and Vehicle Parts Segmentation",
"authors": "Cátia Loureiro, Sandra Dixe, Cristiana Neto, António Linhares Silva",
"venue": "IEEE Access",
"cites": 4,
"link": "https://doi.org/10.1109/access.2026.3653645",
"tasks": "segmentation; detection",
"models": "YOLOv8; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; part localisation + damage; synthetic data / GAN; transformer backbone",
"seed": false,
"abstract": "The growing urbanization of modern societies intensifies the demand for sustainable mobility that aligns efficiency, safety, and environmental responsibility in smart cities. Shared vehicle systems address this need, but remain limited by subjective manual inspections that hinder operational efficiency. Automated exterior damage assessment via computer vision offers a scalable alternative, although its success critically depends on high-quality, domain-specific datasets, which remain scarce. To overcome this limitation, we introduce SYNDCAR (SYNthetic Damage CAR dataset), a generative AI-based dataset that simulates realistic vehicle damage in images with precise annotations for detection and vehicle parts segmentation. Extensive experiments were conducted that covered hyperparameter tuning, transfer learning, benchmarking of CNN-and Transformer-based models, and post-training optimization and quantization. The results highlight the challenges of damage detection due to appearance variability and dataset size, while pointing to practical solutions: YOLOv8-X achieved the highest accuracy, whereas YOLOv12-S provided the best trade-off. TensorRT deployment tests (FP16/INT8) further reduced the inference time by ≈ 71% on the GPU and the energy consumption by 65%. Segmentation remained more energy-intensive than detection. In general, SYNDCAR demonstrates the potential of synthetic datasets to enable accurate, efficient, and scalable vehicle inspection systems."
},
{
"n": 48,
"year": 2022,
"title": "VEHICLE DAMAGED DETECTION USING DEEP LEARNING",
"authors": "",
"venue": "International Research Journal of Modernization in Engineering Technology and Science",
"cites": 3,
"link": "https://doi.org/10.56726/irjmets29626",
"tasks": "detection",
"models": "YOLOv4",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": "Object Detection based Vehicle Damage Detection system can potentially save insurance companies Million.In this paper we propose a Vehicle Damage Detection System based on YOLO v4.We created a dataset of damaged vehicles and annotated regions consisting of Dents, Shattered Glass, damaged tail lights and scratches.We then pre-processed the images inline with the requirements of YOLO v4 and trained the model to achieve an mAP@50 of 81.20.The system proposed in this paper surpasses all previously proposed methods and gives promising results."
},
{
"n": 49,
"year": 2025,
"title": "HL-YOLO: Improving Vehicle Damage Detection with Heterogeneous Convolutions and Large-Kernel Attention",
"authors": "Weijun Li, Huawei Xie, Peiteng Lin, Liyan Huang",
"venue": "World Electric Vehicle Journal",
"cites": 3,
"link": "https://doi.org/10.3390/wevj16120640",
"tasks": "detection",
"models": "YOLOv11",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); lightweight / mobile deployment",
"seed": false,
"abstract": "Accurate vehicle damage detection is essential in intelligent transportation systems, insurance claim assessment, and automotive maintenance. Although conventional detection models demonstrate strong performance, they still struggle to capture fine-grained details and long-range dependencies, which can constrain their effectiveness in real-world applications. To address these limitations, we propose HL-YOLO, an enhanced YOLO11-based architecture that integrates Heterogeneous Convolutions (HetConv) to improve feature extraction diversity and Large-Kernel Attention (LSKA) to strengthen contextual representation. Model evaluation results on a vehicle damage dataset demonstrate that HL-YOLO consistently outperforms the YOLO11 baseline, achieving relative improvements of 2.5% in precision, 5.8% in recall, 3.9% in mAP50, and 3.1% in mAP50–95. These results underscore the model’s robustness in identifying complex damage types, ranging from scratches and dents to accident-induced damage. Although inference latency increased moderately due to the added architectural complexity, the overall accuracy gains confirm the effectiveness of HL-YOLO in scenarios where detection reliability is prioritized over real-time speed. The proposed model shows strong potential for deployment in insurance automation, intelligent traffic monitoring, and vehicle after-service systems, providing a reliable framework for accurate vehicle damage assessment."
},
{
"n": 50,
"year": 2025,
"title": "Enhanced spatial awareness of vision-language model for car damage detection",
"authors": "Yunfei Liu, Quan Long",
"venue": "Journal of Electronic Imaging",
"cites": 1,
"link": "https://doi.org/10.1117/1.jei.34.2.023037",
"tasks": "segmentation; detection; VLM",
"models": "CLIP; VLM/MLLM; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); transfer learning / pre-training; instance segmentation masks; part localisation + damage; vision-language / LLM",
"seed": false,
"abstract": "Car damage detection aims to identify and locate various types of damage on the vehicle, such as dents, scratches, and cracks. Existing works tend to employ deep learning-based object detection approaches for car damage detection, i.e., Mask R-convolutional neural network. With the emergence of large-scale vision-language models (VLMs) such as contrastive language-image pretraining (CLIP), transferring these large-scale models to downstream tasks has become the most popular paradigm in the current field of deep learning. We aim to leverage pre-trained CLIP for car damage detection. However, directly using CLIP is prone to overfitting the training distribution, which is attributed to the absence of robust spatial awareness ability. To this end, we propose to enhance spatial awareness of CLIP from two aspects. First, we propose the spatial-aware feature injection (SFI) that pre-trains a car parts segmentation network to obtain spatial-aware features which are injected into different layers of CLIP’s image encoder through our spatial-aware low-rank cross-attention (SLCA) block. On the contrary, the latent image-text alignment characteristic in CLIP enables us to customize the input for the CLIP’s text encoder. Subsequently, we propose the cross-modal object-contextual representation (COR) module which consists of a cross-modal alignment module to facilitate the alignment of image and text features and an object-contextual refinement module to establish object-contextual represen"
},
{
"n": 51,
"year": 2024,
"title": "CAR DAMAGE DETECTION USING CNN",
"authors": "Rahul Maruti Dhokane",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 1,
"link": "https://doi.org/10.55041/ijsrem30508",
"tasks": "detection; classification; severity",
"models": "Mask R-CNN; VGG",
"datasets": "",
"reported_metrics": "",
"techniques": "instance segmentation masks; severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "In today's modern society, automobiles play a crucial role, and the automatic classification of car damages holds particular significance for the auto insurance industry. Our proposed solution involves the implementation of two Convolutional Neural Network (CNN) models. Specifically, the VGG16 model is employed to identify and assess the location and severity of car damage, while the Mask R-CNN is utilized to accurately mask the damaged regions. Both models collectively provide valuable insights into the extent. The CNN models effectively filter out images without damages, allowing only those with identified damage to be passed on to the object detection model. This strategic approach enhances the overall performance of the model. The core objective of this research project is to achieve maximum accuracy through the utilization of CNN models. TensorFlow, Key Words: E-commerce, Car Damage, Detection, Classification, VGG, Mask RCNN, Severity, Location, Masking"
},
{
"n": 52,
"year": 2023,
"title": "A Comparative Study of Generative Adversarial Networks for Generating Car Damaged Images",
"authors": "Phyu Mar Kyu, Kuntpong Woraratpanya",
"venue": "Research Square",
"cites": 1,
"link": "https://doi.org/10.21203/rs.3.rs-2509412/v1",
"tasks": "classification",
"models": "ViT; Inception; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); data augmentation; transfer learning / pre-training; synthetic data / GAN; damaged-vs-intact pre-filter; transformer backbone",
"seed": false,
"abstract": "Abstract The deeper the deep learning (DL) models are, the more computational complexity in need of the vast amount of data training requires to get better performance with accurate results. Generative adversarial networks (GANs) have obtained tremendous attention from many researchers with their impressive generation of synthetic instance data via a few source data by alleviating the problems of data scarcity, insufficient data diversity, and producing only limited plausible alternative data using standard data augmentation techniques conforming to the art of low-data-driven DL training in both scratch models, and pre-trained model in a variety of image classification tasks. That is why to defeat the above-referred problems and a lack of publicly available high-quality car-damaged datasets in car damage analysis, we created a custom data with a framework including three different evaluation assessments to generate a synthesized car-damaged dataset as the comparative study of Cycle-Consistent Adversarial Networks (CycleGAN), and Attention-Guided Generative Adversarial Network (AttentionGAN) by transforming one domain to another with our custom car damaged-undamaged dataset. In addition to this, we evaluated our generated car-damaged images based on three different evaluation assessments: firstly using three quantitative GANs metrics such as Inception Score (IS), Fréchet Inception Distance (FID), and Kernel Inception Distance (KID); secondly creating a convolutional neural net"
},
{
"n": 53,
"year": 2026,
"title": "Harris Hawks–tuned severity-aware YOLOv8 instance segmentation framework for vehicle damage assessment",
"authors": "Surjeet Dalal, Yogesh Kumar Sharma, Shakti Kundu, Monika Dhananjay Rokade",
"venue": "Scientific Reports",
"cites": 1,
"link": "https://doi.org/10.1038/s41598-026-50925-0",
"tasks": "segmentation; detection; severity",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "",
"techniques": "instance segmentation masks; severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "The severity of vehicle damage is very important in determining road safety, insurance claim and automated vehicle inspection. Although the recent methods of computer vision were promising in the localization of damaged areas, it is difficult to discriminate the level of damage precisely because of subtle visual variations, class disparity, and irregular damage images. In a bid to solve these problems, this paper suggests a severity-sensitive YOLOv8-based instance segmentation system, which incorporates curriculum learning, Harris Hawks Optimization (HHO) to hyperparameter optimization, and confidence thresholding at the per-class level. The proposed learning strategy in the curriculum allows the training to be progressive, whereby initial training occurs with understanding of types of coarse damage, and then further training is given in fine-grained severity discrimination. HHO is used to automatically tune important hyperparameters with severity-sensitive fitness objectives to enhance the robustness of segmentation without changing the architecture. Also, class-specific confidence thresholds are proposed to adjust the precision and recalling to the levels of severity. Experimental analysis of a curated vehicle damage severity dataset shows that convergence is stable and the performance is competitive, with Box mAP50 of 0.271 and Mask mAP50 of 0.135, showing the best performance at an early training phase. The proposed lightweight YOLOv8s-Seg model has shown to be better tha"
},
{
"n": 54,
"year": 2025,
"title": "Deep Learning for Car Damage Detection: A Comprehensive Review",
"authors": "Abhishek Narsayya Bathula, Khan Faiyaz Ahmed, Gaikwad Vaishnavi, Prof. Rashmi Mahajan",
"venue": "International Journal of Advanced Research in Science Communication and Technology",
"cites": 0,
"link": "https://doi.org/10.48175/ijarsct-30380",
"tasks": "detection; severity",
"models": "Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); transfer learning / pre-training; severity / cost estimation; explainability (Grad-CAM)",
"seed": false,
"abstract": "The accurate and efficient assessment of vehicle damage is a critical requirement for industries such as automotive insurance, car rentals, and resale markets. Traditional manual inspection methods are slow, subjective, and prone to inconsistency. With the rapid evolution of Artificial Intelligence, particularly deep learning and computer vision, automated car damage detection has gained significant research attention. This review presents a comprehensive study of major advancements in deep learning-based vehicle damage detection, covering foundational CNN models, transfer learning approaches, object detection frameworks, and hybrid architectures. Key challenges—including limited datasets, difficulty detecting subtle damages, lack of severity estimation, and computational complexity—are highlighted. The review also discusses future research directions to achieve fully automated, robust, and interpretable real-world systems"
},
{
"n": 55,
"year": 2026,
"title": "AI Powered Car Damage Detection Using Deep Learning",
"authors": "Shrujana KC and Prof. Thejaswini MN",
"venue": "International Journal of Advanced Research in Science Communication and Technology",
"cites": 0,
"link": "https://doi.org/10.48175/ijarsct-37886",
"tasks": "detection; classification; severity",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "",
"techniques": "part localisation + damage",
"seed": false,
"abstract": "Damage appraisal of vehicles is a critical component of motor insurance, car repair estimation, used car appraisal, and accident documentation. Manual appraisal methods generally carried humans. The car repair technician inspects the damaged vehicle, detects the damaged parts, and estimates the visual damage extent. Although manual intervention may be necessary in the subsequent stages of legal, structural, or monetary analysis of the damages, the procedure could be both tedious and time-consuming. The proposed project focuses on automating the first stage of this procedure using a computer vision system.The project employs an object detection model using the YOLOv8 framework for detecting and classifying six different kinds of damages of cars"
},
{
"n": 56,
"year": 2025,
"title": "Interactive Deep Learning System for Automated Car Damage Detection: Multi-Model Evaluation and Interactive Web Deployment",
"authors": "Sailaja Madhu, Bharathi Maddikatla, Ranjitha Padakanti, Vineel Sai Kumar Rampally",
"venue": "",
"cites": 0,
"link": "https://doi.org/10.38124/ijisrt/25apr1759",
"tasks": "detection; multi-view/3D",
"models": "YOLOv9",
"datasets": "",
"reported_metrics": "",
"techniques": "data augmentation; lightweight / mobile deployment; 3D / gaussian splatting",
"seed": false,
"abstract": "This project presents an automated framework for vehicle damage evaluation employing deep learning methodologies, designed to optimize assessment procedures within automotive service environments. By implementing the YOLOv9 computational vision architecture, the system enables rapid identification of vehicular damage components through advanced pattern recognition, reducing reliance on labor-intensive manual inspections. The model underwent training on an extensive curated dataset comprising 8,450 annotated images capturing diverse damage morphologies across multiple vehicle perspectives, including frontal collisions, lateral impacts, and rear-end accidents. The framework integrates physics-informed augmentation strategies to enhance environmental adaptability, particularly addressing challenges posed by variable lighting conditions and reflective surfaces. A modular processing pipeline facilitates scalable deployment through quantization techniques optimized for edge computing devices, demonstrating practical applicability in service center operations. The system incorporates a web-based interface enabling real-time damage visualization and automated report generation, significantly streamlining technician workflows. Experimental results indicate substantial improvements in inspection efficiency, with the YOLOv9 architecture achieving 87% mean average precision (mAP@0.5) while maintaining computational efficiency. Quantized model variants exhibited a 68% reduction in memory "
},
{
"n": 57,
"year": 2026,
"title": "Deep Learning Approach for Vehicle Damage Analysis and Repair cost Estimation",
"authors": "Velicheti Srinivas, Garaga Bala Venkata Aditya, Tammisetti Shanmukha Aravind, Vidiyala Lahari",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 0,
"link": "https://doi.org/10.55041/ijsrem58713",
"tasks": "detection; severity; VLM",
"models": "YOLOv5; GPT-4V/4o",
"datasets": "",
"reported_metrics": "",
"techniques": "severity / cost estimation; vision-language / LLM",
"seed": false,
"abstract": "Abstract As more cars hit the roads, accidents and damage are just part of the deal. Usually, experts handle vehicle damage assessments by hand—lots of waiting, high costs, and sometimes you don’t even get the same answer twice. That’s why this paper introduces a deep learning system to automate the whole process, from spotting damage on your car to figuring out what it’ll cost to fix. Here’s how it works: The system pulls in the YOLOv5 object detection model to scan uploaded photos for damage—think dents, scratches, busted parts, or crumpled panels. Then, OpenAI GPT-4o- mini steps in, crunches the damage data, estimates repair costs, and delivers a straightforward explanation. You don’t need any fancy tech skills to use it, either. It’s all packed into a web app, with React.js running the front and Python Flask running the back. KEYWORDS: Vehicle Damage Detection,Deep Learning,YOLOv5,Object Detection,Image processing,Repair Cost Estimation,OpenAI GPT-4o- mini ,Web based System,Artificial Intelligence."
},
{
"n": 58,
"year": 2023,
"title": "Car Dirtiness and Damage Detection For Automatic Service Recommendation Using Machine Learning Techniques",
"authors": "Mohammed Abdullah Khan, Gundlapally Siri Reddy, Tarun Ramavath",
"venue": "International Journal of Scientific Research in Science and Technology",
"cites": 0,
"link": "https://doi.org/10.32628/ijsrst2310115",
"tasks": "detection",
"models": "YOLOv5; VGG; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; lightweight / mobile deployment",
"seed": false,
"abstract": "Automobile Industry is growing by a huge percentage in a few decades contributing about 7.5% to India’s total GDP. As the number of vehicle owners is increasing the demand and need for automobile service is also high but people are busy with their routines hence failing in proper maintenance of their vehicles. In this paper by using machine learning algorithms, and object detection we have come up with the idea to develop a web application that can suggest to users some offers and timing for their car maintenance by analyzing a car using computer vision without the owner’s involvement. The primary Aim of this project is to maintain a vehicle without disturbing the owner’s day-to-day routine. This project is built using the latest technologies and the most trending domains in the industry. We have used the YOLOV5 (You Only Look Once) object detection model and VGG16 architecture to analyze the car images and Flask framework to create a responsive interface for the web application. We generally don’t realize that multiple tasks can be done at a time resulting in many tasks being incomplete one of which is vehicle maintenance. This project aims at both owner’s convenience and the growth of the service provider’s business. In this paper, we present the Machine Learning based automated car maintenance system with effective time utilization which is an IOT device that could be installed at the parking’s main gate of places where people tend to spend many hours like offices or malls"
},
{
"n": 59,
"year": 2025,
"title": "A New Deep Learning Based Damage Classification Pipeline for Car Glass",
"authors": "Ömer Küllü, Anıl Çelik, Pınar Ece Aktan Hatipoğlu, Mehmet Kıvılcım Keleş",
"venue": "Journal of Advanced Research in Natural and Applied Sciences",
"cites": 0,
"link": "https://doi.org/10.28979/jarnas.1651526",
"tasks": "segmentation; detection; classification; severity",
"models": "Ensemble",
"datasets": "",
"reported_metrics": "",
"techniques": "ensemble / model fusion",
"seed": false,
"abstract": "Intelligent damage evaluation is a popular domain of applied Artificial Intelligence (AI) in the insurance industry. Car glass damage is a common insurance claim. Accurately identifying the nature of glass damage is essential, as it directly affects the decision between repair and replacement, which has different safety and cost implications. This distinction is crucial to ensure vehicle safety and effective claim processing. This paper proposes a deep learning (DL) enhanced approach to reduce the claim process time by automatically detecting damage in car glass. The primary objective of this study is to assess whether glass damage is suitable for repair or necessitates total replacement. Owing to variations in the image zoom levels for damaged glass, our method incorporates a branched pathway to handle these differences effectively. Our pipeline employs both classification and segmentation methods. The proposed approach begins by filtering the images. Based on the results, the image is fed directly to the damage classification or subjected to further processing before classification. Furthermore, ensemble and fusion techniques were employed to improve overall efficiency. The ensemble-based strategy achieved an F1 score of 0.99 in damage classification, outperforming other approaches."
},
{
"n": 60,
"year": 2022,
"title": "Damage Detection of An Automobile",
"authors": "R Melvin Raj, R. Jyosthna, Ramya Madhuri N, Akshay Sunny R",
"venue": "International Journal of Engineering Research in Computer Science and Engineering",
"cites": 0,
"link": "https://doi.org/10.36647/ijercse/09.10.art014",
"tasks": "segmentation; detection; severity",
"models": "U-Net; Xception",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "lightweight / mobile deployment",
"seed": false,
"abstract": "As a result of the proliferation of automobile in- dustries today. There have been an increasing number of car accidents, not all of which are serious, but the automobile is damaged. Detecting automobile damage at the site of an accident using images is exceptionally beneficial as it may significantly lower the cost of processing the insurance reimbursement process while also providing more convenience to automobile users. In most cases, this damage is detected and assessed manually from the car’s images during the car evaluation process. In this paper, we worked on the problem of automation of vehicle damage detection which can be used by insurance companies to automate the process of vehicle insurance claims in a rapid fashion. The recent advances in computer vision largely due to the adoption of fast, scalable, and end-to-end trainable Convolutional Neural Networks make it technically feasible to recognize vehicle damages using semantic segmentation. We manually collected and annotated images from various online sources containing different types of vehicle damages and we used U-NET architec- ture to detect the damage of an automobile."
},
{
"n": 61,
"year": 2026,
"title": "Vehicle Damage Detection and Severity Estimation",
"authors": "Sarabu Joshna Sarabu Joshna",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 0,
"link": "https://doi.org/10.55041/ijsrem67459",
"tasks": "detection; classification; severity",
"models": "YOLOv8; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; severity / cost estimation; lightweight / mobile deployment; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Vehicle damage assessment is key to the insurance claim process and for evaluating damaged vehicles.The traditional approach uses manual inspections which are slow, subjective, and often inconsistent due to the many different conditions vehicles are subject to in the real world.This paper proposes an automated system to identify vehicle damage and estimate the severity using deep learning and image processing methods.The authors use a pre-trained YOLOv8 model to find the damage regions, namely dents, scratches, and cracks, using bounding box localization methods.The authors then use the damage zones identified by YOLOv8 to calculate a damage ratio in each zone so that they may be classified as either minor, moderate, or critical damage using a rule-based system.Following this, they have developed a Repair Cost Estimation module to take the severity level of the damage and map it to a predefined repair cost range.Additionally, the authors have used an image hashing method to identify duplicate images submitted and prevent insurance fraud.The findings show that the proposed automated system can successfully identify and classify the severity of vehicle damage and consistently provide accurate cost estimates in real-world conditions.Overall, the proposed automated system provides more efficient vehicle inspection and insurance claims processing by reducing the amount of manual labour and allowing for faster decision making.This work presents several major original contributions:"
},
{
"n": 62,
"year": 2025,
"title": "Vehicle Damage Detection with YOLO and Visual Embedding Fusion",
"authors": "Hyeon-jin Yu, Hee-Cheol Kim",
"venue": "The Journal of the Korean Institute of Information and Communication Engineering",
"cites": 0,
"link": "https://doi.org/10.6109/jkiice.2025.29.11.1481",
"tasks": "detection",
"models": "YOLOv8; Swin; ResNet",
"datasets": "",
"reported_metrics": "",
"techniques": "transformer backbone",
"seed": false,
"abstract": "본 연구는 차량 손상 이미지에서 손상 유형을 자동으로 분류하는 딥러닝 기반 모델을 제안한다. 먼저 YOLOv8 모델을 활용하여 차량 이미지 속 손상 부위를 자동으로 탐지하여 분리한다. 이후 분리된 이미지는 ResNet50, ConvNeXt V2, Swin Transformer 등의 모델에 입력함으로써 서로 다른 시각적 임베딩 벡터를 추출한다. 추출된 임베딩은 결합하여 MLP 분류기에 입력되어 최종적으로 손상 유형을 판별한다. 실험에는 실제 차량 손상 사례가 포함된 AI Hub 데이터셋을 사용하였으며, 그 결과 제안한 Concat 모델의 성능이 단일 모델에 비해 우수하였다. 특히 ResNet50 모델과 비교했을 때 Macro F1-score 기준 약 14%p 향상된 0.74를 기록하였다. 이는 다양한 모델의 시각 정보를 결합하는 접근이 유사한 손상 유형을 구분하는 능력을 향상시킴을 보여준다."
},
{
"n": 63,
"year": 2026,
"title": "Smart-Claim: AI-Based Vehicle Damage Detection and Estimation Using YOLO and Large Language Models",
"authors": "Asan Nainar M, Jeevanandan B K",
"venue": "Iconic Research and Engineering Journals",
"cites": 0,
"link": "https://doi.org/10.64388/irev9i9-1715302",
"tasks": "detection; severity; VLM",
"models": "YOLO",
"datasets": "",
"reported_metrics": "",
"techniques": "vision-language / LLM",
"seed": false,
"abstract": "Vehicle damage inspection is an essential step in insurance claim processing and vehicle repair management. Traditional inspection procedures rely on manual verification by surveyors, which often results in delays, human errors, and inconsistencies in repair estimation. To address these challenges, this research proposes Smart-Claim, an artificial intelligence-based system designed to automate vehicle damage detection and provide repair estimation insights using computer vision and large language models. The system utilizes the YOLO object detection model to identify damaged regions in vehicle images uploaded by users. The detected damage information is analyzed using NVIDIA Nemotron-3 API powered by the Llama 3.1 90B large language model to generate intelligent estimation insights. The application is implemented using React for the frontend, Python Flask for the backend, and MySQL for database management. Experimental results demonstrate that the proposed system significantly reduces inspection time and enhances efficiency in insurance claim evaluation."
},
{
"n": 64,
"year": 2026,
"title": "CarSenseAI: A Deep-Learning Pipeline for Real-Time Vehicle Damage Detection, Severity Assessment, and Repair-Cost Estimation",
"authors": "Mithun R, Pavithra S, Gururaj Katti, Lucky Mahawar",
"venue": "Zenodo (CERN European Organization for Nuclear Research)",
"cites": 0,
"link": "https://doi.org/10.5281/zenodo.22765230",
"tasks": "segmentation; detection; classification; severity",
"models": "YOLOv8; Mask R-CNN",
"datasets": "",
"reported_metrics": "",
"techniques": "data augmentation; instance segmentation masks; severity / cost estimation; lightweight / mobile deployment; synthetic data / GAN; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Vehicle damage assessment is traditionally performed through manual inspection by insurance surveyors, workshop technicians, and fleet-management personnel, a process that is slow, subjective, and difficult to scale as accident volumes and insurance-claim loads grow. We present CarSenseAI, an end-to-end, camera-based mobile system that automates damage detection, localization, severity classification, and repair-cost estimation from ordinary smartphone photographs. We give a formal statement of the assessment task as a composition of four learned maps, and specify each stage by its objective: a YOLOv8 detector trained with a CIoU-based composite loss, a Mask R-CNN segmentation stage supplying pixel-level damage area, a CNN severity classifier over a continuous severity score, and a cost regressor whose output is projected onto ontology-derived bounds so that predictions cannot leave the range domain knowledge admits. Synthetic data augmentation via GANs and diffusion models compensates for the scarcity of labelled examples in underrepresented damage categories. The system is delivered as a Flutter mobile application backed by a Node.js/Flask API and a Firebase/MySQL data layer, with GPS-based mechanic discovery, SOS emergency alerts, and service reminders layered on the core pipeline. A review of 41 studies published between 2019 and 2026 shows that most existing systems address only one stage of the assessment workflow rather than a complete pipeline; CarSenseAI is designed "
},
{
"n": 65,
"year": 2024,
"title": "Improved Accuracy of Vehicle Part Detection and Damage Classification using YOLO Algorithm",
"authors": "Man Zhang, Seung-Soo Shin",
"venue": "Journal of Digital Contents Society",
"cites": 0,
"link": "https://doi.org/10.9728/dcs.2024.25.5.1291",
"tasks": "detection; classification",
"models": "YOLO",
"datasets": "",
"reported_metrics": "",
"techniques": "part localisation + damage",
"seed": false,
"abstract": ""
},
{
"n": 66,
"year": 2026,
"title": "COMPARE ONE-STAGE AND TWO-STAGE SEGMENTATION FOR DETECTING VEHICLE DAMAGE IN AN IMAGE",
"authors": "Nataliya I. Boyko, Yurii В. Bodnar",
"venue": "Collection of Scientific Publications NUS",
"cites": 0,
"link": "https://doi.org/10.15589/znp2026.1(503).2.8",
"tasks": "segmentation; detection; classification; severity",
"models": "Ensemble",
"datasets": "",
"reported_metrics": "",
"techniques": "ensemble / model fusion; severity / cost estimation; lightweight / mobile deployment; reflection / false-positive handling; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Purpose.Development of an intelligent predictive maintenance model for Scania truck Air Pressure Systems (APS) to minimize operational costs under conditions of critical class imbalance and asymmetric error costs.Method.The methodology is based on a comparative analysis of ensemble machine learning algorithms, including Random Forest, XGBoost, and CatBoost.To address data sparsity, a specific imputation strategy was implemented where missing values are treated as informative features.The research utilizes cost-sensitive learning techniques to adapt loss functions to an asymmetric 1:50 cost ratio (false positives vs. false negatives).Furthermore, decision threshold optimization (Threshold Tuning) was applied to shift the model's classification boundary in favor of higher recall for the minority \"failure\" class. Results.Experimental results demonstrate that the CatBoost algorithm provides the highest economic efficiency, achieving a total operating cost of 8,640 units, which is 12.7% lower than XGBoost and 20.8% lower than Random Forest.CatBoost correctly identified almost all critical failures, resulting in only 4 false negatives compared to 9 for XGBoost.A modular software prototype was developed using the FastAPI framework, demonstrating the feasibility of deploying the optimized model into real-time diagnostic environments with low latency.Scientific novelty.The advantage of the CatBoost architecture for industrial data with high imbalance is justified, and a method for dir"
},
{
"n": 67,
"year": 2024,
"title": "Car Damage Detection and Patch-to-Patch Self-supervised Image Alignment",
"authors": "Hanxiao Chen",
"venue": "arXiv (Cornell University)",
"cites": 0,
"link": "https://doi.org/10.48550/arxiv.2403.06674",
"tasks": "detection",
"models": "Mask R-CNN",
"datasets": "",
"reported_metrics": "",
"techniques": "instance segmentation masks; active / semi-supervised learning",
"seed": false,
"abstract": "Most computer vision applications aim to identify pixels in a scene and use them for diverse purposes. One intriguing application is car damage detection for insurance carriers which tends to detect all car damages by comparing both pre-trip and post-trip images, even requiring two components: (i) car damage detection; (ii) image alignment. Firstly, we implemented a Mask R-CNN model to detect car damages on custom images. Whereas for the image alignment section, we especially propose a novel self-supervised Patch-to-Patch SimCLR inspired alignment approach to find perspective transformations between custom pre/post car rental images except for traditional computer vision methods."
},
{
"n": 68,
"year": 2025,
"title": "Vehicle Intelligent Damage Assessment Based on Improved YOLOv8-seg",
"authors": "飞扬 王",
"venue": "Modeling and Simulation",
"cites": 0,
"link": "https://doi.org/10.12677/mos.2025.144365",
"tasks": "segmentation; detection",
"models": "YOLOv8",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); small-object / multi-scale features; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "在中国科技快速发展的大背景下，智能化、自动化逐渐渗透到各个领域。车辆的智能定损研究正在有序推进，传统的定损过程需要进行现场勘察和测量，这一过程繁琐且耗时。而智能定损技术通过分析大量的汽车相关数据和损伤图像，能够在几秒钟内完成对损伤的识别和评估。但目前相关应用泛化性不足，受图片分辨率影响，在小目标及区域分割精确度上仍有较大进步空间。针对这些问题，本文以YOLOv8-seg图像分割算法为基准模型提出了一种精确度高的车辆智能定损模型。通过引入NWD (Normalized Wasserstein Distance)损失函数，强化了模型的小目标识别及分割能力，以及对低分辨率图像的特征提取能力，提升了模型的鲁棒性。提出DLKA (Deformable Large Kernel Attention)可变形大核注意力机制与YOLOv8-seg主干网络C2f模块相融合，实现了在计算复杂度略微增加的前提下，模型分割能力的大幅度提升，强化模型的泛化能力。通过实验验证，本文的改进算法相比原始算法，平均精度上提升了8.1%，在车辆智能定损研究中表现出色。In the context of China’s rapidly developing technology, intelligence and automation are gradually permeating various fields. Research on intelligent vehicle damage assessment is being carried out in an orderly manner. Traditional damage assessment processes require on-site inspections and measurements, which can be cumbersome and time-consuming. Intelligent damage assessment technology analyzes large amounts of car-related data and damage images to identify and evaluate damages within seconds. However, current applications lack generalization capabilities and are affected by image resolution, leaving significant room for improvement in accuracy for small targets and regional segmentation. In response to these issues, this paper proposes a high-precision intelligent vehicle damage assessment model based on the YOLOv8-seg image segmentation algorithm. By introducing the NWD (Normalized Wasserstein Distance) loss function, the model enhances its ability to recognize and segment small targets as well as extract feat"
},
{
"n": 69,
"year": 2026,
"title": "Comparative Analysis of CNN Architectures for Vehicle Accident Damage Classification in Intelligent Transportation Systems (ITS)",
"authors": "Eren Dağlı, Yavuz Selim Taşpınar, Metin Mutlu Aydın, Rıdvan Ertuğrul Yildirim",
"venue": "Applied Sciences",
"cites": 0,
"link": "https://doi.org/10.3390/app16157838",
"tasks": "classification",
"models": "ResNet",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": "Traffic accidents are one of the most significant societal problems worldwide, resulting in loss of life and property. In recent years, powerful solutions for the automatic classification of traffic accidents have been offered by deep learning-based image processing methods. This study comparatively evaluated four convolutional neural network architectures (SqueezeNet, ResNet-18, ResNet-50 and AlexNet) with different depths and levels of complexity using the CADD (Car Accidents and Deformation) dataset, which consists of images of vehicle accidents. The model’s performance was examined in detail using accuracy, precision, recall, the F1 score, a confusion matrix and an ROC–AUC analysis. The ResNet-50 model significantly outperformed all others, achieving 52% validation accuracy and balanced F1-score values (~0.50–0.52). It also achieved particularly high precision (0.70), recall (0.79) and AUC (0.922) values in the Totaled class. In contrast, SqueezeNet exhibited significant class bias and struggled to learn the multi-class structure. The AlexNet and ResNet-18 models showed moderate performance, achieving limited success in terms of discriminability, particularly in the ‘Severe’ class. It should be emphasized that the analysis relies on single-frame static images and a relatively small, predefined dataset; the study is therefore intended as a preliminary comparative benchmark rather than a deployable damage-assessment system. Within this scope, the findings indicate that resi"
},
{
"n": 70,
"year": 2025,
"title": "Research on Damage Detection and Recognition System for Automotive Components Based on Stereo Vision and Deep Learning",
"authors": "Yi Zhou, Haozhe Zhao",
"venue": "International Journal of Mechanical and Electrical Engineering",
"cites": 0,
"link": "https://doi.org/10.62051/ijmee.v6n2.03",
"tasks": "detection",
"models": "YOLOv5; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "data augmentation; transfer learning / pre-training; small-object / multi-scale features; lightweight / mobile deployment",
"seed": false,
"abstract": "Damage detection in automotive components is of paramount importance for ensuring vehicle safety and performance. However, traditional detection methods suffer from significant limitations in both efficiency and accuracy. The recent advancements in deep learning and stereo vision technologies have introduced innovative approaches for intelligent damage detection in complex scenarios. This study proposes a damage detection and recognition system for automotive components that integrates stereo vision and the YOLOv5 deep learning algorithm. The research methodology includes constructing a stereo vision data acquisition platform, conducting image preprocessing and depth information extraction, applying various data augmentation techniques to enhance sample diversity, and leveraging transfer learning and hyperparameter optimization to improve model performance. Experimental results demonstrate that the system exhibits excellent performance in detection accuracy, real-time capability, and adaptability to small objects, effectively identifying diverse types of component damage. This research provides a reliable technological foundation for intelligent detection tasks in complex industrial scenarios, contributing significantly to improving quality control efficiency."
},
{
"n": 71,
"year": 2023,
"title": "MARS: Mask Attention Refinement with Sequential Quadtree Nodes for Car Damage Instance Segmentation",
"authors": "Teerapong Panboonyuen, Naphat Nithisopa, Panin Pienroj, Laphonchai Jirachuphun",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2305.04743v5",
"tasks": "segmentation; detection",
"models": "Mask R-CNN",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); instance segmentation masks; small-object / multi-scale features; transformer backbone",
"seed": false,
"abstract": "Evaluating car damages from misfortune is critical to the car insurance industry. However, the accuracy is still insufficient for real-world applications since the deep learning network is not designed for car damage images as inputs, and its segmented masks are still very coarse. This paper presents MARS (Mask Attention Refinement with Sequential quadtree nodes) for car damage instance segmentation. Our MARS represents self-attention mechanisms to draw global dependencies between the sequential quadtree nodes layer and quadtree transformer to recalibrate channel weights and predict highly accurate instance masks. Our extensive experiments demonstrate that MARS outperforms state-of-the-art (SOTA) instance segmentation methods on three popular benchmarks such as Mask R-CNN [9], PointRend [13], and Mask Transfiner [12], by a large margin of +1.3 maskAP-based R50-FPN backbone and +2.3 maskAP-based R101-FPN backbone on Thai car-damage dataset. Our demos are available at https://github.com/kaopanboonyuen/MARS."
},
{
"n": 72,
"year": 2022,
"title": "Convolutional Neural Networks for vehicle damage detection",
"authors": "Robin Enrico van Ruitenbeek, Sandjai Bhulai",
"venue": "Machine Learning with Applications",
"cites": 47,
"link": "https://doi.org/10.1016/j.mlwa.2022.100332",
"tasks": "detection; classification",
"models": "Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; small-object / multi-scale features; reflection / false-positive handling",
"seed": false,
"abstract": "Vehicle damages are increasingly becoming a liability for shared mobility services. The large number of handovers between drivers demands for an accurate and fast inspection system, which locates small damages and classifies these into the correct damage category. To address this, a damage detection model is developed to locate vehicle damages and classify these into twelve categories. Multiple deep learning algorithms are used, and the effect of different transfer learning and training strategies is evaluated, to optimize the detection performance. The final model, trained on more than 10,000 damage images, is able to accurately detect small damages under various conditions such as water and dirt. A performance evaluation with domain experts shows, that the model achieves comparable performance. In addition, the model is evaluated in a specially designed light street, indicating that strong reflections complicate the detection performance."
},
{
"n": 73,
"year": 2023,
"title": "Vehicle Damage Severity Estimation for Insurance Operations Using In-The-Wild Mobile Images",
"authors": "Dimitrios Mallios, Xiaofei Li, Niall McLaughlin, Jesús Martínez del Rincón",
"venue": "IEEE Access",
"cites": 28,
"link": "https://doi.org/10.1109/access.2023.3299223",
"tasks": "segmentation; detection; severity",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation; lightweight / mobile deployment",
"seed": false,
"abstract": "Following a car accident, an insurance company must assess the level of damage to each vehicle to decide on the compensation paid to the insurance customer. This assessment is usually performed by manual inspection, which is costly and time-consuming. Automatic car damage assessment using image data is an under-addressed problem highly relevant to the insurance industry. Although there have been many attempts at solving particular aspects of this problem, we are unaware of any complete solutions available. In this work, we propose a pipeline that uses photographs of a damaged car, collected by the users from multiple angles, together with structured data about the vehicle, to estimate damage severity following an accident. Our proposed pipeline consists of several computer-vision models for the detection of damage and the determination of its extent. Unlike existing approaches in car damage assessment, we use semantic segmentation to understand which parts of the car are damaged, and to what extent. We then extract computer-vision features, indicating the location and severity of damage to each exterior panel, together with structured data, to arrive at an accurate damage cost estimation. We train and evaluate this model on a large dataset of historical insurance claims with known outcomes, all captured in the wild with mobile-phone hardware."
},
{
"n": 74,
"year": 2024,
"title": "Automatic detection of potholes using VGG-16 pre-trained network and Convolutional Neural Network",
"authors": "Satyabrata Swain, Asis Kumar Tripathy",
"venue": "Heliyon",
"cites": 22,
"link": "https://doi.org/10.1016/j.heliyon.2024.e30957",
"tasks": "detection; classification",
"models": "VGG; Transfer learning",
"datasets": "",
"reported_metrics": "accuracy 97.3%",
"techniques": "transfer learning / pre-training; lightweight / mobile deployment; synthetic data / GAN",
"seed": false,
"abstract": "A self-driving car is necessary to implement traffic intelligence because it can vastly enhance both the safety of driving and the comfort of the driver by adjusting to the circumstances of the road ahead. Road hazards such as potholes can be a big challenge for autonomous vehicles, increasing the risk of crashes and vehicle damage. Real-time identification of road potholes is required to solve this issue. To this end, various approaches have been tried, including notifying the appropriate authorities, utilizing vibration-based sensors, and engaging in three-dimensional laser imaging. Unfortunately, these approaches have several drawbacks, such as large initial expenditures and the possibility of being discovered. Transfer learning is considered a potential answer to the pressing necessity of automating the process of pothole identification. A Convolutional Neural Network (CNN) is constructed to categorize potholes effectively using the VGG-16 pre-trained model as a transfer learning model throughout the training process. A Super-Resolution Generative Adversarial Network (SRGAN) is suggested to enhance the image's overall quality. Experiments conducted with the suggested approach of classifying road potholes revealed a high accuracy rate of 97.3%, and its effectiveness was tested using various criteria. The developed transfer learning technique obtained the best accuracy rate compared to many other deep learning algorithms."
},
{
"n": 75,
"year": 2024,
"title": "Powering AI-driven car damage identification based on VeHIDE dataset",
"authors": "Van-Dung Hoang, Nhan T. Huynh, Nguyen Tran, Khanh Le",
"venue": "Journal of Information and Telecommunication",
"cites": 19,
"link": "https://doi.org/10.1080/24751839.2024.2367387",
"tasks": "segmentation; detection; classification",
"models": "",
"datasets": "VehiDE",
"reported_metrics": "",
"techniques": "instance segmentation masks; lightweight / mobile deployment",
"seed": false,
"abstract": "In the realm of automobile insurance, the imperative of automating car damage evaluation has surged, offering streamlined assessment processes and heightened accuracy. Deep learning techniques have notably influenced vehicle damage assessment, reshaping insurance procedures. However, the primary challenge remains in crafting robust models for damage detection and segmentation. This study presents a novel contribution through the development of the Vehicle Damage Detection (VehiDE) dataset, specifically tailored for comprehensive car damage assessment. The dataset, encompassing 13,945 high-resolution images annotated across eight damage categories, serves as a foundational resource for advancing automated damage identification methodologies. Notably, VehiDE offers support for multiple tasks, including classification, object detection, instance segmentation, and salient object detection, thereby fostering versatile research avenues. Through extensive experimental analysis, including the evaluation of state-of-the-art methodologies on VehiDE, this study not only highlights the dataset's efficacy but also unveils new insights into the challenging nature of car damage assessment. Moreover, the study pioneers the exploration of salient object detection techniques in this domain, showcasing their potential in addressing irregular damage types. By offering VehiDE to the research community, we aim to catalyze advancements in the field of car damage assessment, paving the way for more "
},
{
"n": 76,
"year": 2022,
"title": "Recognizing the Damaged Surface Parts of Cars in the Real Scene Using a Deep Learning Framework",
"authors": "Mahboub Parhizkar, Majid Amirfakhrian",
"venue": "Mathematical Problems in Engineering",
"cites": 17,
"link": "https://doi.org/10.1155/2022/5004129",
"tasks": "detection; severity",
"models": "ResNet; Inception",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": "Automatically recognizing the damaged surface parts of cars can noticeably diminish the cost of processing premium assertion that leads to providing contentment for vehicle users. This recognition task can be conducted using some machine learning (ML) strategies. Deep learning (DL) models as subsets of ML have indicated remarkable potential in object detection and recognition tasks. In this study, an automated recognition of the damaged surface parts of cars in the real scene is suggested that is based on a two-path convolutional neural network (CNN). Our strategy utilizes a ResNet-50 at the beginning of each route to explore low-level features efficiently. Moreover, we proposed new mReLU and inception blocks in each route that are responsible for extracting high-level visual features. The experimental results proved the suggested model obtained high performance in comparison to some state-of-the-art frameworks."
},
{
"n": 77,
"year": 2025,
"title": "Predicting car accident severity in Northwest Ethiopia: a machine learning approach leveraging driver, environmental, and road conditions",
"authors": "Abraham Keffale Mengistu, Andualem Enyew Gedefaw, Nebebe Demis Baykemagn, Agmasie Damtew Walle",
"venue": "Scientific Reports",
"cites": 13,
"link": "https://doi.org/10.1038/s41598-025-08005-2",
"tasks": "detection; severity",
"models": "Ensemble",
"datasets": "",
"reported_metrics": "accuracy 82%",
"techniques": "ensemble / model fusion; severity / cost estimation; synthetic data / GAN; explainability (Grad-CAM)",
"seed": false,
"abstract": "Road traffic accidents (RTAs) in Northwest Ethiopia, a region with a fatality rate of 32.2 per 100,000 residents, pose a critical public health challenge exacerbated by infrastructural deficits and environmental hazards. This study leverages machine learning (ML) to predict accident severity, addressing gaps in localized predictive frameworks for low- and middle-income countries (LMICs). Our study aims to predict the severity of car accidents in Northwest Ethiopia via machine-learning techniques. Using a dataset of 2,000 accidents (2018-2023) from police reports, we integrated driver demographics, behavioral factors (e.g., alcohol use, seatbelt compliance), and environmental conditions (e.g., unpaved roads, weather) in North West Ethiopia. Ten ML models, including Random Forest, XGBoost, and LightGBM, were evaluated after addressing class imbalance via the Synthetic Minority Oversampling Technique (SMOTE). Hyperparameter tuning and Shapley Additive explanations (SHAP) provided model optimization and interpretability. Random Forest outperformed other models, achieving 82% accuracy (AUC-ROC: 0.87) post-tuning. Driver age (mean: 44 years) and environmental factors (e.g., nighttime on unlit roads, rainy conditions) were critical predictors, increasing fatal accident likelihood by 62%. SMOTE improved the accuracy of the outperforming random forest accuracy from 78.6 to 82%. Random Forest exhibited the highest recall (0.82) after optimization, while ensemble methods dominated perfo"
},
{
"n": 78,
"year": 2022,
"title": "Deep Convolutional Neural Networks With Transfer Learning for Automobile Damage Image Classification",
"authors": "Xiaoguang Tian, Henry Han",
"venue": "Journal of Database Management",
"cites": 12,
"link": "https://doi.org/10.4018/jdm.309738",
"tasks": "classification; severity",
"models": "Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "data augmentation; transfer learning / pre-training; lightweight / mobile deployment",
"seed": false,
"abstract": "Deep learning models are more capable of handling large and complex datasets that generally appear in the insurance industry than traditional machine learning models. In this study, transfer learning was employed to build and optimize a simulated automobile damage assessment system. Several classic deep learning methods were applied to extract features from original and augmented automobile damage images. Then, traditional machine learning and cross-validation techniques were applied to train and validate the system. The proposed deep learning model demonstrated advantages over traditional machine learning models regarding features extraction and accuracy. Deep learning approaches fused with logistic regression and support vector machine were found performing as well as those with artificial neural networks under two simulated scenarios. With the proposed method, automobile damage images can be evaluated for insurance adjustment purposes automatically, based on the acquired input. Hence, insurers can automate the claim and adjustment process, thereby achieving cost and time savings."
},
{
"n": 79,
"year": 2024,
"title": "CES: Cost Estimation System for Enhancing the Processing of Car Insurance Claims",
"authors": "Ahmed Shawky Elbhrawy, Mohamed Belal, Mohamed Sameh Hassanein",
"venue": "Journal of Computing and Communication",
"cites": 10,
"link": "https://doi.org/10.21608/jocc.2024.339922",
"tasks": "detection; classification; severity",
"models": "YOLO",
"datasets": "",
"reported_metrics": "precision 78.5%; recall 70.24%",
"techniques": "severity / cost estimation; lightweight / mobile deployment; transformer backbone",
"seed": false,
"abstract": "Damage assessment is crucial in determining insurance reimbursements in the car insurance industry. However, manual inspection is time-consuming and financially costly. Artificial Intelligence (AI) offers a promising automatic damage assessment solution; we propose a Cost Estimation System (CES) for car damage volume level recognition and cost estimation. CES extracts damage estimates from mobile imagery data and combines them with structured customer data to generate accurate cost estimates for insurance purposes. This paper adopts the CRISP-DM (Cross Industry Standard Process for Data Mining) methodology to develop a robust and systematic model. Leveraging AI technology such as the (You Only Look Once) YOLO model and Transformers in image classification while expediting the claims process and mitigating fraud risk. Evaluating CES performance indicates the ability to accurately identify and locate damaged regions in car images, with an average precision of 78.50%, an average recall of 70.24%, and a mean Average Precision (mAP) of 0.66. Resulting in satisfactory performance from the curated dataset of 2508 car photos, which is classified by car body parts, and their inspected damage parts for enhancing cost estimation, productivity, accuracy, and time savings."
},
{
"n": 80,
"year": 2024,
"title": "An Insurtech Platform to Support Claim Management Through the Automatic Detection and Estimation of Car Damage from Pictures",
"authors": "Mohab M. H. Atanasious, Valentina Becchetti, Alessandro Giuseppi, Antonio Pietrabissa",
"venue": "Electronics",
"cites": 9,
"link": "https://doi.org/10.3390/electronics13224333",
"tasks": "detection",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": "Claims management is a complex process through which an insurance company or responsible entity addresses and handles compensation requests from policyholders who have suffered damage or losses. This process entails several stages, including the notification of the claim, damage assessment, settlement of compensation, and, if necessary, dispute resolution. Fair, transparent and timely claims management is crucial for maintaining policyholders’ trust while also limiting the financial impact on the insurer. Technological innovations, such as the use of artificial intelligence and automation, are positively influencing this sector, enabling faster and more effective claims management. This study reports on Insoore AI, an insurtech solution that aims to automate a portion of claims management by integrating a computer vision solution based on some latest developments in deep learning to automatically recognize and localize car damage from user-provided pictures."
},
{
"n": 81,
"year": 2019,
"title": "Preventing Car Damage using CNN and Computer Vision",
"authors": "Dr.Avinash Sharma, Aaditi Verma, Dhananjay Gupta",
"venue": "International Journal of Innovative Technology and Exploring Engineering",
"cites": 8,
"link": "https://doi.org/10.35940/ijitee.a5020.119119",
"tasks": "classification; severity",
"models": "Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; severity / cost estimation",
"seed": false,
"abstract": "This research contain convolutional neural network are used to recognize whether a car on a given image is damage or not, from where it is damage and severity of the damage. Using transfer learning to take advantage of available models that are trained on a more general object recognition task, very satisfactory performance has been achieved, which indicate great opportunities of this approach. Car accidents are stressful and the auto claims process is ripe for disruption. Using computer vision to accurately classify vehicle damage and facilitate claims triage."
},
{
"n": 82,
"year": 2024,
"title": "Enhancing car damage repair cost prediction: Integrating ontology reasoning with regression models",
"authors": "Hamid Ahaggach, Lylia Abrouk A, Eric Lebon",
"venue": "Intelligent Systems with Applications",
"cites": 5,
"link": "https://doi.org/10.1016/j.iswa.2024.200411",
"tasks": "severity",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation",
"seed": false,
"abstract": "The estimation of repair costs for car damage is a critical yet challenging task for insurance companies and repair shops. Accurate and the rapid predictions are essential for providing reliable cost estimates to customers. Traditional methods in this domain face multiple challenges, including manual processes and inaccuracies in repair cost estimation, as outlined in our article. This paper introduces a novel approach that combines regression models with ontology reasoning to enhance the accuracy of car damage repair cost predictions. An Ontology for Car Damage (OCD)1 ,2 has been developed, which is meticulously structured and populated using Named Entity Recognition (NER) and Relation Extraction (RE) techniques. This ontology provides a comprehensive framework for organizing and understanding the complex domain of car damage, capturing essential semantic relationships and variables that significantly influence repair costs. By integrating OCD with seven regression models, such as Random Forest and Decision Tree, we have proposed a hybrid methodology that leverages both structured data and semantic understanding. Our approach not only accounts for typical variables such as the type and severity of damage, and labor costs but also identifies novel features through the use of SWRL (Semantic Web Rule Language) rules, enhancing the model’s predictive capabilities. The performance of our models was evaluated using a substantial real-world dataset comprising over 300,000 records. "
},
{
"n": 83,
"year": 2021,
"title": "Modeling and Application of Neural Networks for Automotive Damage Appraisals",
"authors": "Fred Poon, Yang Zhang, Jonathon Roach, David Josephs",
"venue": "SMU Scholar (Southern Methodist University)",
"cites": 3,
"link": "https://scholar.smu.edu/cgi/viewcontent.cgi?article=1181&context=datasciencereview",
"tasks": "segmentation; detection; classification; severity",
"models": "Mask R-CNN",
"datasets": "",
"reported_metrics": "",
"techniques": "instance segmentation masks",
"seed": false,
"abstract": "The automotive damage appraisal process is one of the areas in property and casualty insurance that can benefit from applying deep learning technology and computer vision. It is commercially beneficial to introduce a fast and efficient claim process that can shorten the entire process. Technologies adopted include advanced neural network algorithm and Mask R-CNN to solve tasks such as image classification, object detection, and segmentation in combination with statistical analysis and model construction of the appraisal metadata to approximate final claim cost. With a database of over 3 million records as the data source, a workflow is constructed via a combination of image modeling and Estimate Management Standard (EMS) modeling which can automate the appraisal process with a one-stop approach. It will increase both the efficiency and the accuracy of the process by reducing the labor cost by 50%, as well as reducing the appraisal timeline from an industry standard five days to just hours on average."
},
{
"n": 84,
"year": 2023,
"title": "Vehicle Insurance Damage Detection",
"authors": "A Jaya Sree, G Aravind, H Lalith, M Yuvraj",
"venue": "International Journal for Research in Applied Science and Engineering Technology",
"cites": 2,
"link": "https://doi.org/10.22214/ijraset.2023.49847",
"tasks": "detection; severity",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation",
"seed": false,
"abstract": "Abstract: These days every process we see has been automated. The Use of Artificial Intelligence and Machine Learning have been adopted by many people and businesses all around the world. One of the adoption is in the sector of vehicles, wherein it can identify the parts that have been damaged of a vehicle with the use of the widespread technologies that are present. The Machine can do many things like predict the repair the damage needs and the cost estimation for the damage. All of this is possible due to the Annotation of Images/Videos of the vehicles that have been damaged and building Computer Vision ML Models. The ML models does everything from detecting, predicting the amount of damage occurred and the cost required to repair it. With the help of these data Insurance Companies can be benefitted by eliminating the scope of fraud in claims processing."
},
{
"n": 85,
"year": 2025,
"title": "Implementing Redefining Vehicle Damage Detection with Deep Learning Revolutionizing Auto Assessments Using Convolutional Neural Network",
"authors": "Md Moshiur Rahman, Arif Khan",
"venue": "Journal of Scientific and Technological Research",
"cites": 0,
"link": "https://doi.org/10.59738/jstr.v6i1.24(70-79).clsg8069",
"tasks": "detection; classification; severity",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "severity / cost estimation; lightweight / mobile deployment; dataset contribution",
"seed": false,
"abstract": "The rising number of road accidents and vehicle collisions has led to the development of advanced systems for assessing and predicting vehicle damage. This paper introduces a computer vision-based approach to predict vehicle damage severity using machine-learning algorithms to analyze images of damaged vehicles (Saxena, 2022). A comprehensive dataset of vehicle damage images with severity labels was collected, and a convolutional neural network (CNN) was trained for damage classification. The CNN model achieved high accuracy in identifying various damage levels. Additionally, the system integrates real-time image processing for seamless integration with surveillance cameras and accident detection systems, facilitating rapid data acquisition and analysis (Ijjina et al., 2019). The system offers automatic damage assessment, benefiting insurance companies, automotive industries, and law enforcement agencies by enabling prompt evaluation of accident scenes. Extensive experiments on a large-scale dataset showed the system's effectiveness, outperforming traditional methods with high precision and recall rates. The proposed system represents a significant advancement in vehicle damage prediction, offering an efficient, reliable, and automated solution for assessing vehicular harm, with potential benefits for road safety infrastructure, insurance processing, and vehicle maintenance (Huang et al., 2020)."
},
{
"n": 86,
"year": 2020,
"title": "Car Exterior Damage Detection Using Mask R-CNN",
"authors": "Pradhip Kumar.R, Sanjeev Patel, S. Shankar",
"venue": "International journal of advance research and innovative ideas in education",
"cites": 0,
"link": "https://openalex.org/W3014118404",
"tasks": "detection; classification",
"models": "Mask R-CNN; Faster R-CNN; Cascade R-CNN; Transfer learning",
"datasets": "",
"reported_metrics": "",
"techniques": "transfer learning / pre-training; instance segmentation masks",
"seed": false,
"abstract": "Recently, the production of image-based vehicle insurance is an important area with considerable scope for automation reach. In this paper we consider the issue of classifying car damage, where some of the categories may be fine-granular. To this reason, we are exploring deep learning-based techniques. Initially we try to train a CNN directly with a collection of training data. However, it's not working well due to a small collection of labeled data. Hence, we investigate the domain-specific pre-training effect accompanied by fine-tuning with a large number of annotated training-data. As Faster R-CNN and SVM have not identified damaged cars with high accuracy, and the Cascade R-CNN takes an immense amount of time to train and check the data that we are working on to fit. Hence, we are training data into a R-CNN Mask that produces adequate results compared to traditional Neural Networks. Though there is a lot of unknowns such as partial images, Hence the classifier was built to detect amorphous damages. The model is layered over 3 classifications of detecting the car and examining whether the damage dealt is high or low. Finally, the classifier is projected with the flask environment to make the working experience easier, since it runs on a localhost the compile time does not exceed 5 seconds irrespective of the quality of the image. Experimental results indicate that Mask R-CNN works better than convolutional R-CNN as transfer learning works better than domain specific fine-t"
},
{
"n": 87,
"year": 2026,
"title": "Vehicle damage detection with transformer-based multi-object tracking",
"authors": "Sergio A. Pérez-Zarate, José Luis Pro-Martín, Juan A. Álvarez-García, Miguel Ángel Martínez del Amor",
"venue": "Engineering Applications of Artificial Intelligence",
"cites": 0,
"link": "https://doi.org/10.1016/j.engappai.2026.114725",
"tasks": "detection",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "lightweight / mobile deployment; reflection / false-positive handling; transformer backbone",
"seed": false,
"abstract": "Vehicle damage assessment requires reliable and consistent localization over time, a challenge that image-based detection systems do not fully address due to their lack of temporal modeling. This paper presents a video-based multi-object tracking framework for automated vehicle damage inspection, built on an adapted real-time detection transformer architecture using a tracking-by-detection paradigm. The proposed approach incorporates a contrastive learning objective to encourage discriminative, identity-consistent embeddings and introduces a transformer-based refinement module to improve temporal coherence across frames under multi-instance damage scenarios. The method is evaluated on a newly collected vehicle damage tracking dataset composed of annotated inspection videos. Experimental results show that the proposed framework improves identity association compared to end-to-end transformer-based tracking baselines, while achieving competitive overall tracking performance relative to detection-driven pipelines. In particular, the model achieves a Higher Order Tracking Accuracy score of 46.79 while maintaining a speed of 21 frames per second, compared to 22.88 obtained by the multiple-object tracking with transformer under the same evaluation protocol, and reaches a multi-object tracking accuracy of 28.94. These findings suggest that incorporating temporal context through structured association mechanisms can enhance tracking stability and reduce false positives in vehicle dam"
},
{
"n": 88,
"year": 2022,
"title": "Car Damage Mask RCNN",
"authors": "M Dhanushree, K. L. Arunkumar",
"venue": "International Journal of Advanced Research in Science Communication and Technology",
"cites": 0,
"link": "https://doi.org/10.48175/ijarsct-5085",
"tasks": "detection",
"models": "Mask R-CNN",
"datasets": "",
"reported_metrics": "",
"techniques": "instance segmentation masks",
"seed": false,
"abstract": "The most challenging task is to detect the vehicle damage using a picture. In this proposed project vehicle damage detection using mask zonal convolutional neural network is to estimate of the damaged area of the car. Using higher technology like computer vision and machine learning the car damage detection with masking can be performed. Consider the situation of the accident place, the user can just send the photo of their damaged car and the company will upload the image to the computer, the computer will show the picture of damaged region with masking. Using higher mechanization like sighting system exp and machine learning will help to noticing the car damage detection with masking."
},
{
"n": 89,
"year": 2024,
"title": "INSURANCE AMOUNT PREDICTION BASED ON ACCIDENTAL CAR DAMAGE LEVEL USING CNN",
"authors": "B Aditya",
"venue": "INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT",
"cites": 0,
"link": "https://doi.org/10.55041/ijsrem33450",
"tasks": "classification",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "lightweight / mobile deployment",
"seed": false,
"abstract": "There is a paradigm shift in the automotive industry with the advancement of technology, especially in the area of ​​insurance claims handling. In this study, we propose a new method for predicting insurance claims based on car damage levels, utilizing advanced artificial intelligence algorithms for image analysis, particularly employing Convolutional Neural Networks (CNN) for image processing and classification. Our strategy is to provide car damage estimates through picture analysis, making insurance claims in the automobile sector simpler. In order to facilitate user interaction with Flask and increase the predictability of our models, we have also implemented enhancements to the user interface. By combining image analysis with a user-friendly interface, our method offers a practical solution for insurance companies to expedite claim processing, reduce fraudulent activities, and ensure fair settlements for policyholders. This research contributes to the ongoing transformation of insurance claim processing within the automotive industry, highlighting the potential of merging image analysis methodologies with user interfaces for enhanced efficiency and usability. Keywords: Insurance claim prediction, Convolutional Neural Networks (CNNs), Image analysis, User interface, Flask, Automotive industry."
},
{
"n": 90,
"year": 2025,
"title": "C-DiffDet+: Fusing Global Scene Context with Generative Denoising for High-Fidelity Car Damage Detection",
"authors": "Abdellah Zakaria Sellam, Ilyes Benaissa, Salah Eddine Bekhouche, Abdenour Hadid",
"venue": "arXiv",
"cites": 0,
"link": "https://arxiv.org/abs/2509.00578v4",
"tasks": "detection",
"models": "",
"datasets": "CarDD",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); synthetic data / GAN",
"seed": false,
"abstract": "Fine-grained object detection in challenging visual domains, such as vehicle damage assessment, presents a formidable challenge even for human experts to resolve reliably. While DiffusionDet has advanced the state-of-the-art through conditional denoising diffusion, its performance remains limited by local feature conditioning in context-dependent scenarios. We address this fundamental limitation by introducing Context-Aware Fusion (CAF), which leverages cross-attention mechanisms to integrate global scene context with local proposal features directly. The global context is generated using a separate dedicated encoder that captures comprehensive environmental information, enabling each object proposal to attend to scene-level understanding. Our framework significantly enhances the generative detection paradigm by enabling each object proposal to attend to comprehensive environmental information. Experimental results demonstrate an improvement over state-of-the-art models on the CarDD benchmark, establishing new performance benchmarks for context-aware object detection in fine-grained domains"
},
{
"n": 91,
"year": 2022,
"title": "Automated vehicle inspection model using a deep learning approach",
"authors": "Mohamed Mostafa Fouad, Karim Malawany, Ahmed Gamil Osman, Hatem Mohamed Amer",
"venue": "Journal of Ambient Intelligence and Humanized Computing",
"cites": 15,
"link": "https://doi.org/10.1007/s12652-022-04105-3",
"tasks": "classification; severity",
"models": "VGG; DenseNet; Ensemble; Transfer learning",
"datasets": "",
"reported_metrics": "accuracy 81%; accuracy 85.5%",
"techniques": "transfer learning / pre-training; ensemble / model fusion; loss redesign (focal/IoU variants)",
"seed": false,
"abstract": "Abstract Image-based inspection is a growing area with a large scope of automation. The automatic classification of vehicle damages would make the insurance claim much faster and more efficient. This can effectively reduce the claiming cost. This paper presents, an image classification model using an adapted version of pre-trained convolutional neural networks. The pre-trained neural networks were, the VGG-19 and DenseNet-169. The proposed model is a pipeline that established with fully connected layers for additional damage classification. The final proposed model improves the feature extraction process. The dataset had a class imbalance problem, so a weighted loss function had been used to solve such problem. The model employed binary cross-entropy as a loss function, and sigmoid activation was applied to the output layers as independent layers. Finally, the model presents a multi-label classifier, where one image may be assigned to many labels. The model classifies vehicle damage through five classes: broken glass, broken headlights, broken taillights, scratches, and dents. A four-layer neural network was employed for the classification, along with several regularization approaches to handle overfitting problem. The final results showed that the DenseNet-169 had a better accuracy of 81%, whereas VGG-19 had a 78%. Another approach had been proposed where it had a mix of transfer and ensemble learning approaches. This final approach had an accuracy of 85.5% and F1-scores of "
},
{
"n": 92,
"year": 2023,
"title": "Car damage assessment recommendation system using neural networks",
"authors": "Jason Elroy Martis, M S Sannidhan, C. V. Aravinda, R. Balasubramani",
"venue": "Materials Today Proceedings",
"cites": 13,
"link": "https://doi.org/10.1016/j.matpr.2023.03.259",
"tasks": "",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": ""
},
{
"n": 93,
"year": 2025,
"title": "YUNet_LLMClaimReport: An Enhanced Automobile Insurance Fraud Detection and Automated Claim Report Generation Using Large Language Models",
"authors": "Priyanka D. Kumar, Sountharrajan Sehar",
"venue": "IEEE Access",
"cites": 2,
"link": "https://doi.org/10.1109/access.2025.3584927",
"tasks": "segmentation; detection; classification; VLM",
"models": "YOLOv11; U-Net",
"datasets": "Kaggle",
"reported_metrics": "",
"techniques": "data augmentation; transfer learning / pre-training; lightweight / mobile deployment; vision-language / LLM",
"seed": false,
"abstract": "The detection of fraud in automobile insurance and processing of automobile claims are still a challenge since manual inspections are carried out and automated systems are fragmented. The state-of-the-art approaches (convolutional neural networks, statistical models, etc.) tend to focus on separate tasks (damage detection, fraud classification, etc.), without being integrated with report generation on an end-to-end basis, and without making sufficient use of unstructured information (e.g., images, claim descriptions). In this research, YUNet_LLMClaimReport, a new framework is proposed that combines YOLOv11, U-Net, and a fine-tuned GPT-3.5-turbo large language model to automatically generate claim reports based on the detections and segmentation. The model is trained on a custom dataset of 4,000 vehicle images within the Egyptian insurance market (70:20:10 training, validation and test split), using data augmentation, dropout (0.3) and L2 regularisation (l = 0.01) to ensure robustness. The evaluation measures are mean average precision (mAP50: 0.9424), intersection over union (IoU: 0.85), claim report accuracy (CRA: 0.92 +/- 0.008), BLEU (0.85 +/- 0.01), and ROUGE-L (0.82 +/- 0.01). The framework runs at 8.3 FPS on NVIDIA Jetson Nano and saves processing time by 60 percent. Weaknesses are single-run measures because of computational limitations and factual inconsistencies in reports (10%). The model can be generalized to the Kaggle Car Damage Detection dataset (mAP50: 0.90), p"
},
{
"n": 94,
"year": 2024,
"title": "VERI-D: A new dataset and method for multi-camera vehicle re-identification of damaged cars under varying lighting conditions",
"authors": "Liu Shao, Sos С. Agaian",
"venue": "APL Machine Learning",
"cites": 2,
"link": "https://doi.org/10.1063/5.0183408",
"tasks": "",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "attention module (CBAM/SE/coordinate); synthetic data / GAN; dataset contribution",
"seed": false,
"abstract": "Vehicle re-identification (V-ReID) is a critical task that aims to match the same vehicle across images from different camera viewpoints. The previous studies have leveraged attribute clues, such as color, model, and license plate, to enhance the V-ReID performance. However, these methods often lack effective interaction between the global–local features and the final V-ReID objective. Moreover, they do not address the challenging issues in real-world scenarios, such as high viewpoint variations, extreme illumination conditions, and car appearance changes (e.g., due to damage or wrong driving). We propose a novel framework to tackle these problems and advance the research in V-ReID, which can handle various types of car appearance changes and achieve robust V-ReID under varying lighting conditions. Our main contributions are as follows: (i) we propose a new Re-ID architecture named global–local self-attention network, which integrates local information into the feature learning process and enhances the feature representation for V-ReID and (ii) we introduce a novel damaged vehicle Re-ID dataset called VERI-D, which is the first publicly available dataset that focuses on this challenging yet practical scenario. The dataset contains both natural and synthetic images of damaged vehicles captured from multiple camera viewpoints and under different lighting conditions. (iii) We conduct extensive experiments on the VERI-D dataset and demonstrate the effectiveness of our approach in"
},
{
"n": 95,
"year": 2024,
"title": "K-Nearest Neighbor Algorithm and Case Base Reasoning on Xenia Car Damage Detection Expert System",
"authors": "Bella Ananda, Raissa Amanda Putri",
"venue": "Journal Of Computer Networks Architecture and High Performance Computing",
"cites": 2,
"link": "https://doi.org/10.47709/cnahpc.v6i2.3700",
"tasks": "detection",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": "PT Astra Daihatsu Motor or commonly abbreviated as ADM is the Sole Agent Brand Holder (ATPM) of Daihatsu cars in Indonesia. Xenia car is one of the most popular cars in Indonesia. Although there are many Xenia car users, it is not uncommon for Xenia cars to experience damage caused by the ignorance of car users, where the car user only knows how to use it but does not know how to maintain the car properly and correctly. Before damage occurs to the car, the car usually experiences several symptoms of damage that the user does not realize. With that, there are often difficulties experienced by users to find out the type of damage to the car. The purpose of applying and designing applications in this study is to apply the Case Based Reasoning method with the K-Nearest Neighbor Algorithm to detect damage to Xenia cars and to design and build applications with the Case Based Reasoning method with the K-Nearest Neighbor Algorithm to detect damage to web-based Xenia cars. This research uses the Research and Development method. Based on the results of research from previous cases, the new case has similarities with 5 cases and the highest similitude value is with the highest type, namely the type of Injector Malfunction damage with a value of 0.625 or around 62.5%. This expert system application can detect and determine the results of Xenia car engine damage detection by applying a method that looks for the closest similarity value of new cases to old cases, namely the Case Based Rea"
},
{
"n": 96,
"year": 2026,
"title": "DeepChiOptimizer for Automated Vehicle Damage Detection Using Hybrid Deep Learning and Feature Selection",
"authors": "Öznur Özaltın",
"venue": "Iranian Journal of Science and Technology Transactions of Electrical Engineering",
"cites": 1,
"link": "https://doi.org/10.1007/s40998-026-01051-0",
"tasks": "detection",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "",
"seed": false,
"abstract": ""
},
{
"n": 97,
"year": 2024,
"title": "Vehicle Damage Analysis",
"authors": "Shubham Bhoite, Suhas Waghe, Aditya Karbhari, Kaushal Salve",
"venue": "International Journal for Research in Applied Science and Engineering Technology",
"cites": 1,
"link": "https://doi.org/10.22214/ijraset.2024.60220",
"tasks": "detection; severity",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "severity / cost estimation",
"seed": false,
"abstract": "Abstract: Vehicle damage, resulting from accidents, environmental factors, and everyday use, presents a ubiquitous challenge to vehicle owners and the broader automotive industry. An efficient and accurate vehicle damage assessment is essential for insurance claims, resale valuations, and overall vehicle safety. However, the traditional damage assessment methods often involve human subjectivity and can be time-consuming, leading to inefficiencies and customer dissatisfaction. This abstract explores the need for modernizing and automating the vehicle damage assessment process. Vehicle damage analysis plays a crucial role in accident reconstruction, insurance claims, safety assessment, and legal proceedings. This abstract provides an overview of the key aspects of vehicle damage analysis, including the methods and tools used, as well as its broader implications. Vehicle damage analysis involves the systematic examination and assessment of damage sustained by vehicles involved in accidents, incidents, or natural disasters. This process is essential for determining the cause and severity of the damage, estimating repair costs, and establishing liability in legal cases. This abstract highlights the multifaceted nature of the problem, encompassing the need for accurate and efficient damage detection, cost estimation, integration with insurance systems, and improved customer experiences. Addressing these challenges requires the application of advanced technologies such as computer v"
},
{
"n": 98,
"year": 2026,
"title": "Integrating LLMs With Computer Vision: A Geometrically Invariant System for First Notice of Loss in Car Insurance",
"authors": "Vito Arconzo, Gerardo Gorga, Gonzalo Gutierrez, Gabriel Pinos",
"venue": "Expert Systems",
"cites": 1,
"link": "https://doi.org/10.1111/exsy.70340",
"tasks": "segmentation; detection; severity; VLM",
"models": "DETR",
"datasets": "CarDD",
"reported_metrics": "",
"techniques": "severity / cost estimation; vision-language / LLM; synthetic data / GAN; transformer backbone",
"seed": false,
"abstract": "ABSTRACT Automating multi‐stage decision workflows in regulated industrial domains demands intelligent systems capable of orchestrating heterogeneous evidence sources, such as uncontrolled photographs, free‐text narratives and structured forms, under strict operational and compliance constraints. We present FADES, a production‐grade expert system that integrates conversational AI with computer vision to automate end‐to‐end claim evidence interpretation and vehicle damage assessment. The system addresses three core challenges: structured data acquisition via state‐machine‐controlled conversational agents powered by tool‐enabled Large Language Models (LLMs); robust damage localization through modern object detection; and deterministic severity estimation via panel‐aware geometric reasoning and rule‐based logic. Our methodology combines backend‐enforced slot‐filling for First Notice of Loss (FNOL) data collection, ensuring compliance with the Italian CAI standard, with a multi‐stage damage analysis pipeline that reconciles detection outputs with vehicle panel segmentation. We systematically evaluated diffusion‐based (DiffusionDet) and transformer‐based (RF‐DETR, D‐FINE) detectors on 13,353 real accident images comprising 41,057 annotations across four damage classes. RF‐DETR Large achieves ( over DiffusionDet), () and , while satisfying real‐time processing requirements. Operational validation demonstrates that the integrated system reduces claim processing time from 30 to 12 mi"
},
{
"n": 99,
"year": 2021,
"title": "Development of The Application for Car Audio Parts Detection Damage Using Case Based Reasoning Method and Nearest Neighbor Algorithm",
"authors": "Andika Saputra, Ali Khumaidi",
"venue": "Jurnal Teknik Informatika C I T Medicom",
"cites": 1,
"link": "https://doi.org/10.35335/cit.vol13.2021.45.pp42-50",
"tasks": "detection",
"models": "",
"datasets": "Own/insurer data",
"reported_metrics": "",
"techniques": "part localisation + damage; lightweight / mobile deployment",
"seed": false,
"abstract": "PT. Denso Ten often receives car audio spare parts that are damaged due to shocks during the trip or sender's error. Damaged parts are collected and repaired by maintenance who has special skills manually. The limited number of maintenance operators and the frequent transfer of experts resulted in work delays due to insufficient spare parts. Spare parts repair work cannot be done by all employees because it requires special skills. The Case-based Reasoning approach and Nearest Neighbor algorithm are used to be developed for expert systems to support the detection of audio part damage so that it will speed up work and can be done by employees without special knowledge. The system can run and be used by users properly as needed and the results have good accuracy. The Case Base Reasoning method and the nearest neighbor algorithm work according to the rules and the calculation results are according to the expert's results."
},
{
"n": 100,
"year": 2026,
"title": "Intact and Damaged Car Parts Detection Using Deep Learning Algorithms Enhanced with Fuzzy Logic",
"authors": "Serge Jonathan Kogbeti-Wongo, Usef Faghihi, Fadel Touré, Axel-Frank Tankam N.",
"venue": "SSRN Electronic Journal",
"cites": 0,
"link": "https://doi.org/10.2139/ssrn.7111687",
"tasks": "detection",
"models": "",
"datasets": "",
"reported_metrics": "",
"techniques": "part localisation + damage",
"seed": false,
"abstract": ""
}
]