Advancements in Real-Time Safety Gear Recognition at Construction Sites: A Deep Learning Perspective

This paper introduces a deep learning methodology for automated safety gear recognition at construction sites, addressing the limitations of labor-intensive and error-prone traditional monitoring methods. Employing a Convolutional Neural Network (CNN), the system is trained on diverse images of construction workers with varying safety gear compliance levels. The real-time model effectively identifies essential safety gear such as helmets, vests, gloves, and goggles, exhibiting high accuracy and low false negatives. Preliminary results underscore the potential for enhanced site safety, accident reduction, and improved regulatory compliance. Beyond construction, this system holds broader applicability across industries requiring safety gear, marking a significant contribution to the role of artificial intelligence in workplace safety. The model achieves a notable precision rate of 90.5%, a recall rate of 88.7%, an F1-score of 89.6%, and an average Average Precision (mAP) of 87.2% at a 0.5 Intersection over Union (IoU) threshold.