A YOLO-POWERED COMPUTER VISION APPROACH TO HELMET DETECTION FOR ENHANCING CONSTRUCTION SITE SAFETY
TL;DR
An advanced “SmartSafety” system that employs computer vision technology, utilizing the cutting-edge YOLO (You Only Look Once) version 12 (YOLOv12) for real-time detection of helmets at construction sites, achieves an average mAP@0.5 accuracy exceeding 94%, effectively distinguishing individuals wearing helmets.
Abstract
Ensuring the safety of workers is of utmost importance in construction management, with helmet compliance serving as a crucial preventive measure against head injuries. This paper introduces an advanced “SmartSafety” system that employs computer vision technology, utilizing the cutting-edge YOLO (You Only Look Once) version 12 (YOLOv12) for real-time detection of helmets at construction sites. By analyzing high-resolution video footage from strategically positioned cameras, our deep learning model achieves an average mAP@0.5 accuracy exceeding 94%, effectively distinguishing individuals wearing helmets. The model's effectiveness is underscored by a consistent decrease in loss and enhancements in training metrics. Experimental results under diverse environmental conditions, including varying lighting and dynamic worker movements, further illustrate the system’s robustness. Beyond fostering compliance with safety regulations, this system encourages a proactive safety culture and opens avenues for scalable applications in occupational health management. Our findings underscore the transformative potential of computer vision technologies in enhancing safety and intelligence within construction environments.