Identity-Based Proactive Human Intrusion Management in Hazardous Areas at Construction Sites: A Deep Learning–Based Method

Abstract

Construction sites are dynamic, complex, and contain many potential hazards. Workers are often focused on their work and therefore have a reduced ability to detect hazards, which results in the frequent exposure of workers to hazardous areas. The conventional approach to safety management relies on manual monitoring, which is inherently inefficient, subjective, and unsuitable for continuous monitoring of large sites. The advancement of computer vision technology has yielded new solutions for monitoring intruders. Nevertheless, the current monitoring methods fail to consider worker identification, and the evaluation rules for intrusions are unduly straightforward. To solve this problem, this study presents a novel identity-based worker intrusion detection method. Firstly, the automated extraction of worker gait silhouette images was achieved using the YOLOv5 object detection model and the portrait segmentation algorithm. Secondly, the GaitSet model for gait recognition was improved to facilitate the real-time recognition of workers’ identities in monitoring videos. Subsequently, the safety rules for assessing the risk of worker intrusion were established by considering the location and interaction time between workers and the hazardous area, as well as the inherent attributes of the hazardous area. Finally, the performance of the proposed method was evaluated on a construction site. The results demonstrated that the automated gait silhouette extraction method employed in this study exhibited superior accuracy in comparison to other methods, such as the Gaussian mixture model. The improved GaitSet model achieved an average rank-1 of 99.11% with high recognition accuracy. The method of this study can be effectively applied in construction sites with an accuracy of 84.69% for worker identification. The proposed method has the potential to facilitate the automatic monitoring of hazardous area intrusion behavior among workers, which could prove beneficial in enhancing the safety management of construction sites.