The increasing demand for precise and real-time analysis of athletic movements has driven the adoption of advanced deep-learning techniques in sports training. Traditional methods often rely on manual observation or shallow machine learning models, which cannot capture complex spatial and temporal dynamics of human motion. Hence, the research proposes a Deep Dynamic Graph Attention Posture Recognition (DDGAPR), a novel deep learning model that integrates graph neural networks with transformer-based attention mechanisms to model the intricate relationships among body joints over time. The module begins with dynamically representing an athlete's skeleton as a graph, applying attention to both spatial and temporal features to improve posture classification accuracy and robustness. In comparison to previous models, the suggested DDGAPR model improves upon them in several ways: motion recognition accuracy by 18%, validation accuracy by 9.4%, true positive classes by 9.1%, and average attention weight by 10%. The research findings suggest that the DDGAPR can enhance training quality, personalize performance optimization, and support accurate posture recognition. The research contributes to the growing body of literature on AI-assisted sports training, opening new pathways for intelligent, data-driven sports development.
Shi et al. (Wed,) studied this question.