Applying YOLOv8 deep learning model to control personal protective equipment usage behavior on construction sites

Abstract

The supervision of workers’ use of PPE is of utmost importance to prevent worksite accidents. The development of image processing, computer vision, and machine learning algorithms has allowed achievements in such safety procedures. This study proposes a method that applies the YOLOv8 deep learning model – one of the most advanced algorithms – to automatically detect the use of safety protective equipment by workers on construction sites, including behaviors of both using and not using safety protective equipment. The research results show significant improvements compared to previous models, specifically in accuracy rates for detecting helmets, gloves, and masks, which reach 96.6%, 99.6%, and 98.2% respectively. Moreover, the model also holds considerable significance in detecting instances of non-compliance with PPE, cases that require increased attention in occupational safety monitoring. Therefore, the proposed deep learning model is capable of practical application on construction sites, enhancing management capabilities and control over PPE usage, thereby ensuring occupational safety for workers during construction activities.