基于计算机视觉的电力作业人员行为分析研究现状与展望  

Research Status and Development of Computer-vision-based Power Workers' Behavior Analysis

作　　者：闫云凤[1,2] 陈汐 金浩远 齐冬莲[1,2] 储海东 汪金维 YAN Yunfeng;CHEN Xi;JIN Haoyuan;QI Donglian;CHU Haidong;WANG Jinwei(College of Electrical Engineering,Zhejiang University,Hangzhou 310027,China;Hainan Institute of Zhejiang University,Sanya 572025,China;Department of Computer Science,The University of Hong Kong,Hong Kong 999077,China;State Grid Corporation of China,Beijing 100031,China;Zhejiang Tianheng Wu Wei Electronics Technology Co.,Ltd.,Tianheng 310027,China)

机构地区：[1]浙江大学电气工程学院,杭州310027 [2]浙江大学海南研究院,三亚572025 [3]香港大学计算机科学学院,中国香港999077 [4]国家电网有限公司,北京100031 [5]浙江天衡五维电子科技有限公司,天衡310027

出　　处：《高电压技术》2024年第5期1842-1854,共13页High Voltage Engineering

基　　金：国家自然科学基金(62127803);浙江省科技计划项目(2022C01056)。

摘　　要：电力作业人员的有效监管是保障电力安全生产的基础。该文对电力视频中作业人员的行为识别研究进行了归类总结,涵盖静态行为分析(穿戴分析、动作分析和组合分析)和动态行为分析(复杂动作、时序行为和行为预测等);详细综述了电力作业行为分析中的核心算法模块,包括目标检测、姿态估计和视频跟踪等;论述了电力作业行为识别在算法高效性、鲁棒性、灵活性等方面所面临的应用难点和挑战,并展望了电力作业行为智能监控领域的未来发展方向,特别强调了在软硬件结合、通用大模型、生成式人工智能方面进行技术创新和改进所蕴含的潜在机会。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.

关 键 词：行为分析 视觉理解 电力监控 目标检测 姿态估计 视频跟踪 行为预测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM08[自动化与计算机技术—计算机科学与技术]