Multi-scenario helmet detection based on improved YOLOv8
Sun Yufei, Wanjun Yu · 2024
With the rapid development of industrial automation and construction industry, for a variety of high-risk scenarios such as high-altitude working areas, power chemical industry and tunnel underground operation, workers’ safety helmet wearing detection is prone to complex environmental interference, small worker targets and scattered distribution, and prone to false detection and missed detection. A PA-YOLO helmet wear detection model based on YOLOv8 is proposed. PA-YOLO first adds a micro-scale detection head on the basis of the original YOLOv8n model, so that it can detect and locate small targets more keenly, and optimize the problem of missing detection and false detection caused by small targets occupying fewer pixels in the image. Secondly, the scale sequence feature fusion module and channel location attention mechanism are introduced to enhance the network’s ability to extract multi-scale information and improve the model’s ability to recognize complex scenes. In the training test on the data set SHWD, mAP0.5and mAP0.5:0.95reach 91.88% and 65.49% respectively, and the reference number is reduced by about 17.15%, which can meet the needs of helmet wearing detection in different industrial and construction scenarios.