Video Behavior Detection based on Optimized Alphapose in Electricity Facility Management
Yuan Tian, Ning Li, Bo Mao · IEEE/WIC/ACM International Conference on Web Intelligence · 2021
Along with the development of IoT (Internet of Things) Technologies, increasing number of cameras are deployed in electric power facilities. To better utilize the video information from these devices, a behavior detection framework is proposed to identify the potential dangerous or damages in electric power facilities based on the human posture extraction algorithm. The proposed method is an optimization from the popular Alphapose. We create the structure vector and its angle to describe the human body. Then the behavior semantic detection is implemented based on the proposed descriptors. The typical dangerous actions in electricity facilities such as smoking, falling, fighting and etc. are detected in a test dataset. The experiment results indicate the proposed framework can effectively identify the typical dangerous behaviors in a reasonable time.