Abstract: In underground coal mine work sites, moving people and objects may appear between the drill pipes and the monitoring camera, resulting in incomplete video footage and counting omissions of drill pipes. At present, studies on drill pipe counting methods based on image processing and machine vision rarely address the problem of occlusion. Most existing models require collecting and processing all frames of the target video and performing image preprocessing. To address the above issues, a drill pipe counting algorithm for underground drilling rigs based on miner operation pose recognition named the BlazePose-DPC algorithm, was proposed. This algorithm used the BlazePose network to extract key pose information of miners as the basis for automatic drill pipe counting, transforming the drill pipe counting task into the recognition and matching of key operational poses of miners. Key poses were extracted as skeletal joint coordinates from key pose frames via the BlazePose network. Key pose coordinate matching used normalized Euclidean distance to represent the similarity between poses. When the similarity exceeded a predefined threshold, the action in the video was considered complete, and the count was incremented by one, thereby enabling automatic drill pipe counting. Experiments on the BlazePose-DPC algorithm were conducted using two datasets. Dataset 1 was recorded by a mobile device at the Qinggangping Coal Mine in Xunyi, Shaanxi Province, where video instability was common. Dataset 2 was recorded by a fixed surveillance device at the Huaneng Qingyang Meidian Hetaoyu Coal Mine, where uneven lighting and occlusion were common. Experimental results showed that the BlazePose-DPC algorithm was able to perform accurate counting even under challenging lighting conditions or partial occlusion. It maintained accurate counting during prolonged operation, demonstrating stable performance. The BlazePose-DPC algorithm achieved an accuracy of 95.5%, meeting the requirements for drill pipe counting.