Detection and tracking of safety helmet based on DeepSort and YOLOv5

Safety helmets can effectively prevent accidental head injuries to construction personnel. Helmet wearing detection is of great significance to the safety management of the construction site. The existing detection algorithms are difficult to detect small targets and dense targets, and the target occlusion and complex and changeable construction environment will reduce the detection accuracy. To solve the above problems, an intelligent helmet recognition system is designed, which combines multi-target tracking algorithm DeepSort and YOLOv5 detector. Firstly, the target bounding box is extracted by YOLOv5 algorithm and input into DeepSort framework. Further, the target trajectory prediction and tracking are realized by Kalman filter and Hungarian Algorithm. The actual test results on the complex construction sites show that the helmet recognition system based on YOLOv5 and DeepSort improves the detection speed and accuracy compared with the single detection algorithm and partial tracking algorithm. The average accuracy of the system is 94.5%, and the detection speed can reach 40fps, basically realizing real-time detection and providing an effective guarantee for the helmet-wearing detection task.