Detection of construction worker safety protection equipment based on fast YOLO

Abstract

Enterprises are often concerned about disasters, particularly accidents involving personnel. Therefore, there is ongoing research dedicated to safeguarding the safety of employees within enterprises. In recent years, with the continuous development of machine vision technology, industry and academia have been working to adopt machine vision methods to address safety hazards in workers' production. Machine vision applications are still being researched for specific sectors and lack generalization. Algorithms commonly face issues of high computational complexity and demanding hardware requirements. This paper adopts the lightweight YOLOV5 as the baseline algorithm and enhances its accuracy using a receptive field attention mechanism. SeNet is introduced to improve the generalization of object detection, and IDetect Head is employed to increase the efficiency of the detection head. Ultimately, the algorithm's accuracy is enhanced by 3.7%, and mAP50 is increased by 3.0%. This algorithm can be deployed to Internet of Things (IoT) machine vision terminals, reducing deployment costs and improving monitoring efficiency.