Abstract: With the gradual depletion of shallow coal resources, coal mining operations are moving to greater depths. This increases the methane content of coal seams and significantly raises the risk of gas-related disasters. Numerous studies have shown that unsafe behaviors of underground workers are one of the major causes of frequent coal mine accidents. However, current behavior supervision in coal mines still relies on manual video monitoring. This approach suffers from low levels of automation, poor recognition accuracy, and high resource consumption, making it difficult to meet the requirements of real-time safety management during underground operations.To address the challenges of complex drill rod replacement, poor lighting conditions, and difficulties in behavior recognition in underground mines, we developed an integrated detection and counting system for drill rod operations. The system combines object detection, object tracking, pose estimation, and action recognition. Specifically, the YOLO11 model is employed to achieve high-precision object detection. A lightweight ByteTrack algorithm is introduced to improve tracking stability. YOLO-Pose is used to extract human skeletal keypoints, which are then analyzed with a Spatio-Temporal Graph Convolutional Network (ST-GCN) for action classification.Experimental results on a self-constructed dataset show that YOLO11 achieves an F1-score of 93.64%. On two public datasets, the F1-scores reach 67.89% and 63.10%, outperforming other YOLO series models. In object tracking tasks, the system achieves a Multiple Object Tracking Accuracy (MOTA) of 70.7% and an IDF1 score of 65.8%, both higher than those obtained with DeepSORT. For rod-unloading action recognition, the adoption of a spatial partitioning strategy increases accuracy to 87.5%, representing an improvement of 4.17% over conventional methods. In addition, semantic-based judgment enables automatic drill rod counting with an accuracy of 95%.These results demonstrate that the proposed system exhibits strong robustness and real-time performance under conditions of low illumination, heavy occlusion, and complex backgrounds in underground coal mines. The system provides effective technical support for intelligent monitoring and safety management of drill rod operations.