Practical Approaches to Integrating Smart Technology into Physical Education in Higher Education

Abstract

The deep integration of artificial intelligence technology into the physical education curriculum in higher education institutions has significantly accelerated the transition of physical education toward intelligent systems. This paper employs the OpenPose human pose estimation algorithm to extract skeletal joint points from sports videos, and preprocesses the extracted skeletal data by addressing missing and outlier points using mean imputation and exponential smoothing. Based on this, an ST-GCN network model—fusing graph-convolutional neural networks with spatiotemporal attention mechanisms—was constructed to classify sports movements, and the identified movements were evaluated using the DTW algorithm. The results show that the average recognition accuracy of the proposed method for sports movements on the training and test sets was 90.67% and 88.42%, respectively. The model features a significantly reduced number of parameters and offers real-time processing capabilities, outperforming traditional methods. Furthermore, the feedback provided through motion evaluation effectively helps users improve motion quality and meet standard requirements, thereby comprehensively enhancing the quality and efficiency of physical education instruction. This study aims to provide theoretical references and practical approaches for leveraging artificial intelligence to facilitate the systematic and scientific development of higher education teaching systems.