Fall detection of personnel working in deep pits of power transmission lines based on human pose estimation and transformer
Jun Xu, Lei Qian, Peilun Zhang, Jinlong Qi, Min Xie · 2024
To enhance construction safety, we propose a real-time monitoring system using deep learning for detecting falls in complex pit environments. To address this issue, we propose a novel approach that combines human body pose estimation and a Transformer-based method for detecting personnel falls in deep foundation pit operations. Firstly, we extract the key point poses of the human body from the video footage to eliminate interference from complex background information. Then, we encode these estimated key point poses into pose vectors across consecutive frames and use a Transformer network to model the correlation between the sequence of key point poses and personnel fall events. This approach overcomes the limitations of traditional methods that rely on high-quality and large quantities of labeled data, thereby enhancing the ability to detect personnel falls in the deep foundation pit operations of transmission lines. We validate the effectiveness of our proposed method at various deep pit operation sites for transmission lines. The method accurately identifies abnormal events of personnel falls, enabling above-ground managers to promptly perceive the safety conditions of underground operators and take accurate rescue measures. Moreover, our method has potential applications in the power industry and other similar operation scenarios.