Dress Code Surveillance at Power Grid Construction Site via Object Detection

Abstract

Constructions at the power grid working site have potential high risks if the construction rules are not well obeyed. The dress codes at the power grid working site are critical rules to protect the workers' safety. The most fundamental dress code for the workers to follow is to wear the uniform and helmet when entering the working site and wear the safety rope for the aerial operation. However, the negligence of workers or the lack of surveillance can lead to irregular dress situations and may further result in serious accidents. In this paper, with the help of deep learning technology, we proposed to establish a surveillance system for the dress code of power grid workers via object detection. Specifically, we borrow YOLOv5 as the backbone detection model. To detect the dress of the power grid workers, we construct a dataset that contains the common uniform, helmet, and safety belt to fine-tune the pre-trained YOLOv5 model. Experimental results verify the feasibility of our method to provide real-time surveillance of the workers' dress code on the power grid working site.