Monitoring the Abnormal Human Behaviors in Substations based on Probabilistic Behaviours Prediction and YOLO-V5
Yaohui Xiao, Yufeng Wang, Weiming Li, Meng Sun, Xiaojie Shen, Zhengyang Luo · 2022 7th Asia Conference on Power and Electrical Engineering (ACPEE) · 2022
To ensure the safety of substation personnel, unsafe behaviors including not wearing safety helmet, entering dangerous area or smoking are prohibited by the security regulations. Recently, with the large-scale installation of substation video monitoring equipment, it is promising to find the anomalies through automatic image object detection methods. Therefore, this paper presents a security monitoring system for the detection of abnormal behaviors in substations. The system inputs the video monitoring and builds upon a regression deep convolutional neural network for predicting the probability of abnormal behaviors. Based on YOLO-V5 algorithms, the deep neural network is trained by minimizing the multi-part loss function considering locating, sizing, classifying and the probability of abnormal behaviors. The experiment shows that the proposed method can confidently and automatically monitor the abnormal behaviors of people in the substations, locating each person and wearing of helmets, which validates the application potentials of the proposed method under complex environments in substations.