Classification and recognition method of dangerous behaviors of electric power operators based on improved OpenPose algorithm

To achieve intelligent identification of dangerous behaviors of electric power workers in complex environment, a classification and identification method based on improved OpenPose algorithm is proposed. The GSP-Darknet network is used to enhance the extraction of key points of small bones, and the missing joint coordinates are filled in by the average values of adjacent frames. Additionally, a spatiotemporal graph convolution model integrating a graph attention mechanism is constructed to analyze the spatiotemporal characteristics. The experiments show that this method can effectively fill in the missing joints, focus on key information, and accurately identify six kinds of dangerous behaviors, with an accuracy rate of 95.3% and a F1 score of 91%, which is superior to the comparison method and provides a reliable scheme for power safety monitoring.