Detection and Recognition of Personal Protection Equipment Wearing Based on an Improved YOLOv5 Algorithm
E Wanlin, Zhaoxia Yang, Jiaxuan Yu · 2023
An improved YOLOv5 algorithm for small target detection and recognition imposed in this paper, which was deployed in an embedded platform to complete testing experiments. Our experimental evidence shows that the enhanced YOLOv5 algorithm improves detection efficiency. This paper adopts the improved YOLOv5s algorithm to detect and recognize whether workers on site are wearing helmets, masks, or reflective clothing during their operations, ensuring safty. The final model achieves an mAP50 of 0.847, a computational load of 8GFlops, and a model size of 7.4mb. The inference latency on the Ubuntu operating system with an i7-7850cpu is approximately 60ms. Applying the improved YOLOv5s algorithm in the field of construction site safety enables effective monitoring of the proper wearing of helmets and masks by workers, meeting the requirements of real-world applications in terms of detection speed and the performance requirements of portable devices.