Sports Action Detection and Counting Algorithm Based on Pose Estimation and Its Application in Physical Education Teaching

Abstract

Accurate motion detection and counting are crucial for improving training effectiveness and preventing sports injuries in physical education teaching and training. Traditional analysis of sports movements mainly relies on the observation of coaches and the subjective feelings of athletes. This method is not only time-consuming and labor-intensive, but also susceptible to personal experience and judgment biases. To address this issue, a 3D bone keypoint detection algorithm based on the pose estimation model Visual Background Extractor was proposed. By analyzing the continuous action information on the time series, each repeated action was divided and scored using a penalty function. Meanwhile, based on OpenPose, its lightweight application was proposed. The performance test confirmed that the detection accuracy of the playground and outdoor space was the same, both at 96%, which was the highest among all environments. From an individual performance perspective, although the height, gender, and clothing of the participants varied, these factors did not significantly affect the performance of the algorithm, with accuracy ranging from 92% to 94%. These experiments confirm that the proposed motion detection and counting model based on pose estimation has high robustness and reliability under different environmental conditions.