Portal de Periódicos da CAPES

Car Damage Detection and Cost Evaluation Using MASK R-CNN

2021; Springer International Publishing; Linguagem: Inglês

10.1007/978-981-16-3153-5_31

2367-3370

J. D. Dorathi Jayaseeli, Greeta Kavitha Jayaraj, Mehaa Kanakarajan, D. Malathi,

Autonomous Vehicle Technology and Safety

Detecting the damage on a car is an image-based processing method with enormous scope for automation. This concept of automated detection of the extent of exterior damage on a car and subsequent quantification of the damage severity would benefit car insurers, car rentals and repair services. In this paper, we propose employing convolution neural networks to build a Mask R-CNN model that can detect the area of damage on a car. The dataset used consists of images of damaged vehicles with a single class named scratch. The images are precisely annotated with the area of damage. The model is trained using the base weights of Mask R-CNN COCO dataset. The images are processed for 21 epochs. After processing, the final result is visualized using a color splash technique, wherein the area of damage is highlighted. This model would help in reducing the cost of processing insurance claims and lead to greater customer satisfaction. Car dealers can eliminate the manual process of damage assessment and the labor cost accompanied by it. Accuracy and transparency in pricing cars and their potential repairs will be made more prevalent. Fraudulent vehicle insurance claims can also be diminished. On implementing our model, we achieved an overall loss of 0.3888.