A YOLO-v4-Based Risk Detection Method for Power High Voltage Operation Scene

Xinyue Xiang, Feng Zhao, Boya Peng, Hao Qiu, Zhenhua Tan, Zhikang Shuai

Abstract: Power accidents may occur during the period of power operation, which threatens the safety of power operators. It is important to supervise the power operation scene and the safety of workers. Traditional monitoring method for power operation relies on camera and manual operation, which is tedious, labour-intensive and error-prone. To solve this problem, this paper proposes a risk detection method of highvoltage power operation scene based on YOLO v4. Firstly, it selects a specific object “hand-held high-voltage joystick” as the identification object of the high-voltage operation scene. Then the index Degree of Intersection (DoI) is defined to describe the degree of overlap between two bounding boxes. Finally, a safety function is constructed as a risk indicator according to safety requirements. The effectiveness of the proposed method is verified based on tensorflow 12.0 and TITAN RTX platform. The model mAP value reached 91.26%, and the detection speed reached 13f/s, which greatly helps the real-time safety monitoring of the power operation field.