Abstract: Effective supervision of power workers is the basis for ensuring the safe power production. In this paper, we categorize and summarize the research on the behavior recognition of power workers in video, including static behavior analysis (clothing analysis, action analysis, and combination analysis) and dynamic behavior analysis (complex actions, temporal behavior, and behavior prediction, etc.). We provide a detailed overview of the core algorithmic modules in the analysis of power operation behaviors, including target detection, pose estimation, and video tracking, and others. We also discuss the challenges and difficulties in applying these techniques in terms of algorithm efficiency, robustness, and flexibility, and put forward the future development direction of the field of intelligent monitoring of power operation behaviors; meanwhile, the emphasis is put on the potential opportunities in technological innovation and improvement through the integration of hardware and software, general large models, and generative artificial intelligence.