Acceso usuarios
Ouyang Zhaozhang, Jeffrey S. Sarmiento, Zhang Weixiong
To accurately detect motorcycle helmet wearing in complex traffic, an enhanced YOLOV5 method is proposed. The YOLOV5 layers’ functions are analyzed. A dataset of motorcycle helmet wearing in complex traffic environments was constructed using web crawlers and roadside photography techniques. After using AI Studio annotation tools to enhance data processing, a helmet dataset was built using web crawlers and roadside photography. Integrating MobileNetv3, CBAM, and SiLU, the optimized YOLOV5 achieved high detection accuracy in diverse conditions (extreme weather, main roads, heavy traffic). The experimental results show that the designed method can effectively detect the wearing of motorcycle helmets in various complex traffic environments, and the confidence of the generated predicted bounding boxes is greater than 0.9, with a maximum of 0.99. The detection accuracy is 0.954, the recall rate is 0.946, and the average accuracy is 0.969.
