Harness-Wearing Detection of Construction Workers Based on Deep Learning

The death and injury rate of the construction industry is higher than the average level of other industries, and falls from heights account for a large share of the accidents. The automatic monitor of the harness-wearing condition of construction workers can alleviate this problem, but the traditional method such as wearing sensor equipment has many disadvantages, and previous research which used the computer vision methods rarely discussed the automatic monitor of harness-wearing under a specific dangerous scene. In this research, we attempted to analyze the effect of the automatic monitor of the harness-wearing condition using the latest computer vision technology and the feasibility of applying it in a specific scene. First, we set a scene in construction that the construction workers working on the mobile lifting platform (mlp) are detected to need to wear a harness, and we created a dataset about the worker, mlp, and harness for this research. Then we used an objects detection algorithm (YOLOv5) as a technical tool for experimental study, which showed that the mAP of the model was greater than 0.97, and the detection speed was between 9 ms/fps and 15 ms/fps, which met the real-time detection needs in a construction site. Besides, we added conditional detection to detect whether the worker needs to wear a harness and whether they are wearing a harness based on the position relation output on the images. The research in this paper presents a method to detect harness-wearing automatically in a specific scene of construction and shows that applying computer vision technology in specific construction activities has been feasible and valuable.