Research on Real-Time Athlete Tracking Method Based on Computer Deep Learning

With the continuous development of computer deep learning technology, the research of athlete target detection and tracking has attracted more and more attention. The challenge of athlete tracking lies in the changes of sports environment and athletes' clothing. In recent years, with the development of deep learning technology, especially the continuous improvement of target detection and tracking algorithms, the performance of athlete target detection and tracking has been greatly improved. yolov5 object detection algorithm and DeepSORT multi-object tracking algorithm based on object detection have achieved excellent results in the field of pedestrian tracking, but the related research in the field of sports object tracking is relatively limited. This paper adopts yolov5. As the detection part of the athlete tracking algorithm, the attention mechanism is improved, and then the DeepSORT algorithm is improved to realize the real-time tracking of athletes. In this paper, ShuffleNet is used as the apparent feature extraction network to solve the mismatching problem of DeepSORT algorithm in the tracking process, and ECA attention mechanism is added to optimize the feature extraction capability of the network. In the aspect of trajectory matching, DIOU (Distance-IOU) is used to improve the matching process by considering the degree of overlap of target frames and the Distance between target frames. Experimental tests on self-made data sets show that Multiple Object Tracking Accuracy (MOTA) is increased by 3.7% and Multiple ObjectTracking Precision (MOTP) is increased by 2.6%. The results show that the improved method can effectively improve the track disappearance and target ID switching of athletes in the process of target tracking.