Detection of Workers Without the Helments in Videos Based on YOLO V3

Abstract

In the construction site with complex environment, there are a variety of potential risk factors threatening personal safety. Since the head is the most critical part of a human body and is the most vulnerable to fatal injuries, lots of accidents caused by workers not wearing safety helmets have occurred from time to time. So, the staff working in construction site has the requirement to wear the safety helmet. In order to reduce the incidence of safety accidents caused by not wearing a protective helmet, a realtime detection application for wearing helmet based on YOLO (You Look Only Once) v3 was proposed and encapsulated into a real-time detection software with alert function and simple efficient operation, which has been successfully deployed to several construction sites. YOLO v3 model has a good response to the identification of ąśpesonąs; Therefore, firstly, the workers in video are identified and intercepted by YOLO v3 model, and positive and negative samples are made. Then the worker in the dataset is identified whether wearing helmets or not. Compared with y-OLO v3, the mAP (Mean Average Precision) reached 93.5%, and the detection rate reached 35fps. Theoretical analysis and experimental results show that the proposed algorithm not only satisfies the helmet wearing detection task in real time, but also has high detection accuracy.