In this paper, a recognition algorithm that can automate the car insurance claim process that can predict and recognize the amount of car damage and traffic accident types based on a mask region-based convolutional neural network (Mask R-CNN) is proposed. Mask R-CNN is an extended instance segmentation framework of Faster R-CNN, in the first step it scans the image to generate proposals and in the second step it classifies the proposals to generate bounding boxes, classes, and masks.
The proposed algorithm recognizes the amount of vehicle damage and the type of traffic accident using the damaged body part detection strategy. The vehicle damage recognition feature allows you to fairly determine the cost of repairs based on the level of vehicle damage, and the type of accident recognition helps you to determine the proportion of liability for traffic accident issues fairly.
Vehicle damage recognition and traffic accident type recognition performance are performed through the dataset and Mask R-CNN framework. The average accuracy of vehicle damage recognition at all levels was 73%, and the average accuracy of traffic accident type recognition in all collisions was 86%. Based on this study, we plan to propose an algorithm that can better analyze the types of traffic accidents by detecting vehicle routes in the future.