Vehicle Damage Detection Using Machine Learning and Segmentation to Aid Insurance Claims

The manual assessment of vehicle damage is time consuming, prone to bias and inaccuracies, compromising the validity of insurance claims. To ensure precise, efficient, and timely claim processing, automated evaluation of vehicle damage is essential. However, the auto insurance industry lacks a comprehensive solution that integrates classification, object detection, and segmentation techniques for accurate and thorough damage analysis. This research aims to address this challenge by developing a machine learning pipeline with six models. The pipeline focuses on image-based input and automates the identification and assessment of vehicle damage. VGG16, VGG19, ResNet50, Inceptionv3, and Custom CNN models are compared to classify damaged cars. For Object detection YOLOv5, YOLOv6, YOLOv7, and YOLOv8 models are evaluated to detect specific damaged parts. For segmentation tasks, SegFormer, UNet, and YOLOv8 Instance segmentation models are compared to detect the specific type of damage. The most accurate models for each task are selected through rigorous evaluation. The proposed pipeline achieves a real-time test accuracy of 92.00%. This not only enhances customer satisfaction but also improves the evaluation process within the auto insurance industry.