Research On The Key Frame Selection Method Of YOLOv5-Based Intelligent Damage Determination System

In vehicle damage claims, the intelligent damage determination method based on computer vision instead of manual observation has become the future direction of technology. In the process of wise damage determination, because of the difficulty that the images taken manually cannot be quickly shot with the image quality applicable to the system, this paper proposes a keyframe selection method based on the YOLOv5 model, which solves the problems of convenience and ease of use of the intelligent damage determination system by adding the real-time operation guidance function at the time of the shooting. Firstly, the mobile real-time detection model YOLOv5 detects and counts the vehicle appearance parts. Secondly, the underlying logic determination model guides the user to complete the shooting distance adjustment. Experimenting with the self-built vehicle image data set, the average accuracy of mAP50 reached 94.10/0, and YOLOv5 can meet the requirements of real-time and accuracy of keyframe selection in shooting environments such as uneven lighting and complex surrounding background.