MatchingDPC: Drill Pipes Counting Based on Matching Key Pose Encoding

Underground coal mines face the threat of gas explosions, where drill pipes play a crucial role in reducing natural gas concentration, directly impacting coal mine safety. In this study, Matching DPC model, which simplifies and enhances the accuracy of drill pipes counting in underground coal mines is proposed by employing encoding and matching technology. The model comprises two modules, the pose encoding network and the similarity matching network. First, we extract certain frames from the complete cyclic actions as key pose frames which can represent miner actions. Then, the key pose frames are passed into the pose encoding network to obtain the key pose encodings. Finally, the similarity between the key pose encoding and the video stream is calculated by the similarity matching network, evaluating whether the drill pipe operation actions are completed. With the above approach, we simplify drill pipe counting into a similarity matching task. Unlike the complex design required for detecting drill pipes or rig in RGB video data models, MatchingDPC’s design, which is concise and compact, adheres to the principle of Occam’s razor. This model maintains a low parameter count by utilizing key pose data, simultaneously eliminating interference from complex backgrounds such as uneven lighting, obstruction and motion blur in underground coal mines. The model could be more flexible in capturing actions of different complexities by adjusting the number of key pose frames. Experiments demonstrate that MatchingDPC exhibits better robustness and we achieve a recognition accuracy of 96.0% on the real-word dataset. Additionally, the model’s optimization parameters are merely 2.3M, and the model size is just 10.8MB. Above results verify the practicality of the method.