Abstract
Construction safety accidents have emerged in an endless stream in recent years, causing many casualties and property losses. Among them are caused by insufficient supervision on the construction sites and workers’ low safety awareness. However, the traditional manual management method which consumes manpower and resources is no longer applicable. Thus, in this study, a novel single-stage model based on YOLOv8s was suggested, mainly used for two purposes: for one thing, workers’ personal protective equipment detection, and for another, monitoring and recognizing whether workers enter dangerous areas, feeding back the detection results in real-time to reduce the occurrence of construction accidents. Besides, a brief design of distance calculation was proposed. The model was trained for 200 iterations on a dataset originating from Roboflow, consisting of 103,500 annotated images. According to experiment results, YOLOv8s outperformed YOLOv8n, YOLOv5s, and YOLOv5n in terms of detection performance. It achieved mAP50 of 84%, precision of 85%, and recall of 60.5% in nine detection classes. With the aid of artificial intelligence technology, this study seeks to provide an effective method for managing construction site safety, which might be improved even further with additional images and a more robust network architecture.