Improved detection network model based on YOLOv5 for warning safety in construction sites

Abstract

The safety of worker guarantee is a crucial task in construction site management. Many accidents occur in construction sites by falling, collisions, electrocutions, or being stuck in operating devices. The suitable personal protective equipment (PPE) stated in safety rules is widely used to ensure workers’ safety. The use of PPE is relied on traditional methods such as physical monitoring and video observation that waste time, poor timeliness, and missed inspections. To overcome these limitations, this study utilized newly You Only Look Once (YOLO) algorithm, named YOLOv5, which includes four network structures, namely YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x for safety detection. A data set with 11978 samples was used to establish a digital safety monitoring system via training and testing phases. The comparison results among the four models show that the YOLOv5s performed the best and the average detection speed reached 110 frames per second, which fulfils the real-time detection requirements. This study contributes to the state of the knowledge by (i) providing a one-step solution for the automatic identification the PPE on construction sites; (ii) proposing a valuable tool to assist site safety engineer in the task of automatically detecting the PPE worn by construction workers; and (iii) the effectiveness and superiority of the presented approach are demonstrated via large detection dataset with 11978 samples and real construction case.