Safety helmet detection based on improved YOLOv7-tiny with multiple feature enhancement
Shuqiang Wang,Peiyang Wu,Qingqing Wu
DOI: https://doi.org/10.1007/s11554-024-01501-0
IF: 2.293
2024-06-26
Journal of Real-Time Image Processing
Abstract:Safety helmets are vital protective gear for construction workers, effectively reducing head injuries and safeguarding lives. By identification of safety helmet usage, workers' unsafe behaviors can be detected and corrected in a timely manner, reducing the possibility of accidents. Target detection methods based on computer vision can achieve fast and accurate detection regarding the wearing habits of safety helmets of workers. In this study, we propose a real-time construction-site helmet detection algorithm that improves YOLOv7-tiny to address the problems associated with automatically identifying construction-site helmets. First, the Efficient Multi-scale Attention (EMA) module is introduced at the trunk to capture the detailed information; here, the model is more focused on training to recognize the helmet-related target features. Second, the detection head is replaced with a self-attentive Dynamic Head (DyHead) for stronger feature representation. Finally, Wise-IoU (WIoU) with a dynamic nonmonotonic focusing mechanism is used as a loss function to improve the model's ability to manage the situation of mutual occlusion between workers and enhance the detection performance. The experimental results show that the improved YOLOv7-tiny algorithm model yields 3.3, 1.5, and 5.6% improvements in the evaluation of indices of mAP@0.5, precision, and recall, respectively, while maintaining its lightweight features; this enables more accurate detection with a suitable detection speed and is more in conjunction with the needs of on-site-automated detection.
computer science, artificial intelligence,engineering, electrical & electronic,imaging science & photographic technology