A NOVEL MSF-GGCN MODEL FOR 3D IMAGE ANALYSIS OF LONG-DISTANCE RUNNERS

This paper aims to provide a detailed framework for improving long-distance runners’ performance. Accordingly, leveraging the foundation of artificial intelligence (AI) and the fundamental principles underpinning graph neural network (GNN), this paper delves into the examination of both recursive graph neural network (RGNN) and graph convolutional neural network (GCNN). Additionally, the spatial convolutional neural network (SCNN) and frequency convolutional neural network (FCNN) are introduced to address the spatial structure and frequency domain characteristics inherent in the joint motion data of long-distance runners effectively. Within this context, a novel approach termed the Multi-Stream Fusion-Gated Graph Convolutional Network (MSF-GGCN) is proposed. Founded on the construction of a human skeleton diagram, this method scrutinizes potential dependencies among key points at the terminal nodes of the skeleton. It enhances the ability to distinguish similar actions by analyzing movement data from joints. Experimental findings reveal that MSF-GGCN enhances recognition accuracy in Kinetics-Skeleton datasets by 5.8% and 6.5%, respectively, compared to the Spatial-Temporal Graph Convective Network (ST-GCN). This paper improves how accurately runners’ actions can be recognized using multi-modal fusion, helping to better their training and injury recovery.