Personalized physical education teaching interventionmodel in colleges based on Spatiotemporal GraphConvolutional Network for 3D action recognition

Abstract

Abstract Physical education (PE) in colleges requires precise and personalized instructional feedback, which is difficult to achieve through traditional subjective observation. While 3D skeleton-based action recognition offers a promising alternative, existing Spatiotemporal Graph Convolutional Networks (STGCNs) rely on fixed physical anatomical graphs, failing to capture complex, non-adjacent joint synergies in intense sports. Furthermore, they lack the capability to evaluate asynchronous student movements for fine-grained biomechanical error detection. To address these limitations, this study proposes an end-to-end personalized teaching intervention framework based on an Adaptive Spatiotemporal Graph Convolutional Network (A-STGCN). The A-STGCN dynamically learns data-driven topological structures via a self-attention mechanism, decoupling feature aggregation from rigid physical assumptions to effectively model long-range joint correlations. To bridge the gap between visual recognition and pedagogical evaluation, we integrate a constrained multidimensional Dynamic Time Warping (DTW) algorithm. This module aligns student execution sequences with expert templates, enabling precise spatial error localization regardless of individual rhythmic variations. Extensive experiments demonstrate that the A-STGCN achieves state-of-the-art performance, reaching a Top-1 accuracy of 92.5% on the custom PE-Action dataset and outperforming the strong baseline CTR-GCN by a significant margin of 3.3%. More importantly, a real-world pedagogical experiment confirms the practical significance of the intervention system: students receiving real-time, rule-based corrective feedback achieved a significantly higher post-test average score (88.5) compared to the control group (78.2). This framework provides a reliable and interpretable algorithmic foundation for intelligent, individualized modern physical education.