Detection of wearing safety helmet for workers based on YOLOv4

Abstract

As a necessary protective equipment for workers to enter and exit the construction environment, safety helmets are of great significance to ensure the safe operation of workers. However, there are still some workers who lack safety awareness and do not wear safety helmets from time to time, and there are great safety hazards. This paper is based on the target detection algorithm of YOLOv4, focusing on the real construction site, and real-time detection of workers' helmet wearing in complex scenes. In order to solve the common phenomenon that only one type of helmet is detected, the helmet that is standing on the table or held in the hand is also recognized as a worker wearing a helmet. This article adds a human body model based on the helmet training. Training makes the detected helmet and the human body have a one-to-one correspondence. Experimental results show that the model achieves 93% accuracy on 9986 hard hat data sets. At the same time, the model has been deployed to the actual construction site to meet the daily detection of workers' hard hats, which verifies the effectiveness of the algorithm.