An Improved Merge-YOLOv5 Algorithm Applied to Helmet Wear Detection

Safety management is a vital link in site operation, and wearing a safety helmet is crucial in protecting workers' heads in the construction environment. However, some site operators need more safety awareness, and accidents caused by not wearing safety helmets often occur. Therefore, it is of great significance to monitor the wearing of safety helmets. This paper attempts to improve the traditional target detection algorithm YOLOv5 by updating the NMS function in the original algorithm, which improves the recognition accuracy of the algorithm by making the prediction frame more accurate. After the improvement, compared with the original YOLOv5, the performance indexes of Soft-YOLOv5, Ciou-YOLOv5, Diou-YOLOv5, Giou-YOLOv5, and Merge-YOLOv5 are improved. The [email protected]:0.95 performance indexes of Merge-YOLOv5, Soft-YOLOv5, Ciou-YOLOv5, Diou-YOLOv5, and Giou-YOLOv5 increased by 3.0%, -1.4%, 2.5%, 2.3%, and 2.4% compared with original YOLOv5. The results show that the improved performance of Merge-YOLOv5 is outstanding, and it is more suitable for the on-site inspection of safety helmets.