CIB-SE-YOLOv8: Optimized YOLOv8 for Real-Time Safety Equipment Detection on Construction Sites

Safety equipment is a crucial component in ensuring safety on construction sites, with helmets being particularly essential in reducing injuries and fatalities. Traditional methods, such as manual checks by superintendents and project managers, are inefficient and labor-intensive, often failing to prevent incidents related to the lack of safety gear. To address this, a new approach leveraging computer vision and deep learning has been developed, utilizing real-time detection techniques, specifically YOLO. This study utilizes the publicly available SHEL5K dataset for helmet detection tasks. The proposed CIB-SE-YOLOv8 model builds upon YOLOv8n by incorporating SE attention mechanisms and replacing certain C2f blocks with C2fCIB blocks. Compared to YOLOv8n, our model achieved a mAP50 of 88.4%, representing a 3.2% improvement. It also demonstrated a 0.5% increase in precision and a 3.9% increase in recall, significantly enhancing helmet detection performance. Additionally, the proposed model, with 2.68 million parameters and 7.6 GFLOPs, offers greater efficiency in real-time detection tasks compared to YOLOv8n’s 3 million parameters and 8.1 GFLOPs, making it a more effective solution for improving construction site safety.