Abstract: In order to solve the problems of low accuracy and poor real-time performance of existing target tracking algorithms in the complex environment of coal mines, a YOLO-FasterNet+ByteTrack coal mine personnel tracking algorithm was proposed based on the Tracking by Detection (TBD) paradigm. Firstly, the FasterNet-Block feature extraction module was constructed to improve the Backbone of YOLOv7 and improve the real-time performance of the object detection stage. Then, the CBAM attention mechanism was introduced into Neck to improve the feature perception ability of the model in complex scenes. Then, Soft-NMS is introduced in the decoding stage of object detection to optimize the detection accuracy of the model in personnel overlapping scenario. Finally, in the target tracking stage, a multi-target motion feature prediction mechanism fused with GRU and Kalman filter was designed to solve the problem of target ID flipping caused by personnel overlap and occlusion, which effectively improved the accuracy of coal mine personnel tracking. Experimental results show that the average accuracy of YOLO-FasterNet is increased by 3.6% and the detection speed is increased by 8.2FPS compared with YOLOv7 on the coal mine personnel dataset, and the MOTA value of the proposed target tracking algorithm is increased by 1.7% and the IDSW is reduced by 149 times compared with ByteTrack on the custom tracking dataset GBMOT.