Dangerous Action Recognition Method for High Climbing Workers Based on Industrial Internet of Things

High-altitude operations are ubiquitous in various industrial scenarios, such as construction, petrochemical, and the electric power industry. Due to the high-risk factors in high-altitude climbing, the safety of workers can be improved by using the intelligent surveillance system to recognize dangerous actions in real-time. In this paper, the dangerous action recognition method is proposed for high climbing workers based on computer vision technologies. Considering real industrial environments, computer vision technologies are used to estimate the worker posture even the worker is obscured by industrial objects, and the transfer learning technology is used to enhance the model performance. Based on safety regulations and deep learning, the action analytical model is proposed to recognize the worker dangerous actions automatically. Experiments show that the proposed method performs well in recognizing dangerous actions of the worker in the video.