Drill Pipe Counting Method in Coal Mine Based on Improved Object Tracking
Xiaojun Wu, Peiru Xie, Sheng Yuan, Qiao Hu, Xinyi Wang, Yue She · 2023
With the advancement of mining intelligence, the application of artificial intelligence in this field is increasing. However, how to accurately count the number of drill pipes and improve safety production is still a challenge. To address this issue, one algorithm is proposed that involves analyzing the drilling pattern and using visual object tracking to count drill pipes. The You Only Look Once (YOLO) is improved by using rotated object detection with the circular smoothing label to obtain the position and angle information of the drill in the first frame. A dynamic template update strategy is adopted to improve the accuracy of Transformer Tracking (TransT) in long-term object tracking. Finally, the drill pipe is counted by using the peak points of the trajectory and worker actions with PoseConv3D. Compared with the baseline, the proposed algorithm achieves an average accuracy of 95.62% and a counting efficiency of 42.5 frames per second (FPS), making it more robust and accurate in the field of drill pipe counting.