PPE Patrol: YOLO To The Rescue! – Advanced Object Detection For Enhanced Safety Compliance

Abstract

In spite of various safety precautions, the construction sector still encounters more fatalities in comparison to the other industries. While Personal Protective Equipment (PPE) is designed to prevent/mitigate any accidents, workers often overlook its use, whether unintentionally or on purpose. Manually monitoring safety compliance is challenging due to the large workforce, though ensuring their safety remains a top priority for site managers. Performing manual checking of safety mechanisms is troublesome because of a large number of workers available on-site but maintaining their security is topmost in site authorities' considerations. We conceived an automatic PPE recognition system based on computer vision (CV) technology to tackle this problem. The system is capable of recognizing various categories of PPE. As a part of this study [1], we also proposed a new dataset named CHVG, containing four colored hardhats, vests, safety glasses, person bodies, and person heads, with a total of eight distinct classes. Approximately 2100 images constitute the dataset that we have utilized, and each of them containing details of these eight classes labeled on it. We employed the use of the YOLOv8, which is well known for being anchor-free in its architecture, for the detection process. YOLOv8 is faster and more accurate than existing object detection models. The possibility of employing cutting-edge computer vision techniques to enhance safety monitoring activities in hazardous settings, such as construction sites, is demonstrated by this study