AUTHOR=Zhu Jiaju , Ye Zijun , Ren Meixue , Ma Guodong 

TITLE=Transformative skeletal motion analysis: optimization of exercise training and injury prevention through graph neural networks

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 18 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1353257

DOI=10.3389/fnins.2024.1353257

ISSN=1662-453X

ABSTRACT=In today's society, exercise has become a key pathway to maintaining physical health. However, incorrect postures and movements can lead to sports injuries, making skeletal motion analysis a crucial field for improving exercise effectiveness and reducing potential harm. This research aims to integrate advanced technologies such as Transformer, Graph Neural Networks, and Generative Adversarial Networks to optimize sports training and enhance injury prevention. We focus on key metrics such as specificity, accuracy, recall, and F1-score to comprehensively evaluate the performance of the proposed method.Firstly, we model skeletal motion sequences using a Transformer network to capture global correlation information. Subsequently, we use a Graph Neural Network to model local motion features for a deeper understanding of the relationships between joints. To enhance the model's robustness and adaptability, we introduce a Generative Adversarial Network, employing adversarial training to generate more realistic and diverse motion sequences.In the experimental validation phase, we utilize skeletal motion data sets from multiple groups, including professional athletes and regular fitness enthusiasts. Compared to traditional methods, our approach shows significant improvements in specificity, accuracy, recall, and F1-score. Specifically, specificity increases by approximately 5%, accuracy reaches around 90%, recall improves to around 91%, and the F1-score exceeds 89%. Overall, our proposed skeletal motion analysis method based on Transformer and Graph Neural Networks achieves notable success in optimizing exercise training and preventing injuries. By effectively utilizing global and local information and incorporating Generative Adversarial Networks, our method excels in capturing motion features and improving precision and adaptability. In the future, we will continue 1 Sample et al.in-depth research to advance this field and provide more reliable technological support for healthy exercise.