Safety Helmet Detection: A Comparative Analysis Using YOLOv4, YOLOv5, and YOLOv7

Abstract

Safety helmets are of utmost importance to workers’ lives as the most fundamental form of protection. However, safety helmets are frequently not worn as a result of a lack of safety awareness. Utilizing outdated manual inspection techniques and video monitoring to check if employees are wearing helmets results in missed inspections and poor punctuality. As object detection technologies advanced, the YOLO family of detection algorithms, which have extremely high speed and precision, were applied in multiple detection segments. In this paper, we compare and analyze the three models of the YOLO family, the YOLOv4, the YOLOv5, and the YOLOv7 for helmet detection. A publicly available dataset of 5000 images was collected and annotated. Our results have shown that the YOLOv7 accomplishes an mAP of 96.4% which is 1.36% better than the YOLOv5 and 3.00% better than the YOLOv4. The results also show that the YOLOv7 has an average detection time of 12.4 ms, outperforming that of the YOLOv4 and the YOLOv5. Both in terms of accuracy and speed, the YOLOv7 exceeds both models, making it possible for even greater real-time object detection.