Research on Safety Helmet Detection Algorithm Based on Improved YOLOv5s.

Area of Science:

• Computer Vision and Deep Learning
• Occupational Safety and Health Technology

Background:

• Manual supervision for safety helmet compliance is inefficient and unreliable in high-risk workplaces.
• Existing object detection algorithms struggle with precision, especially for small targets like safety helmets.
• Automated helmet detection systems are needed to enhance safety and enforce compliance.

Purpose of the Study:

• To develop a highly accurate and efficient deep learning model for safety helmet detection.
• To improve upon the YOLOv5s network for enhanced performance in real-world workplace scenarios.
• To enable real-time monitoring of safety helmet usage.

Main Methods:

• A modified YOLOv5s network incorporating Global Attention Mechanism (GAM) and Convolutional Block Attention Module (CBAM).
• Enhancements include recalculating prediction frames, IoU-based clustering, K-means++ anchor modification, and SIoU loss function.
• Knowledge distillation was employed for model lightweighting to achieve real-time detection capabilities.

Main Results:

• The proposed model demonstrated superior performance over the original YOLOv5s in precision, recall, and mean average precision (mAP).
• Achieved more effective identification of safety helmet usage in low-light conditions and varying distances.
• The lightweight design facilitated improved detection speed for real-time monitoring applications.

Conclusions:

• The enhanced YOLOv5s model offers a robust and accurate solution for automated safety helmet detection.
• This technology can significantly improve workplace safety compliance and reduce injury risks.
• The model's effectiveness in challenging conditions makes it suitable for diverse industrial environments.