Establish a Dynamic Detection System for Metal Bicycle Frame Defects Based on YOLO Object Detection

Abstract

The purpose of this research is to develop a real-time bicycle frame's defect detection system using YOLO (You Only Look Once) and machine vision. Firstly, the defect locations are manually selected and a database is established. Next, a Darknet method is used to train the YOLO model. Its static detection accuracy rate is 92.6%, and then the static training model is combined with a robotic arm and an industrial camera to perform dynamic detection verification. The result shows that its detection rate reaches 87%. Finally, the above-mentioned defect detection technology is used with the detection machine to complete the development of the online defect detection system.