Scholarlink

<p>In this research work, a Deep Learning (DL) approach utilizing Spatial Transformer, Temporal Transformer, and Collaborative Movement Centric (CMC) module is presented to classify sports event based on athlete movements in competitive sports. Primarily, the Spatial Transformer utilizing the Spatial Feature extraction module (SFEM) is introduced to extract detailed spatial information from video frames. Here, the SFEM employs Modulated Moving Average Graph Convolutional Network (MMA-GCN) to extract complex spatial relationships by learning the knowledge from offset and modulation parameters. Secondly, the Temporal Transformer is designed that employs Temporal Feature extraction module (TFEM) module to extract long-range temporal dependencies and model evolution across consecutive frames. Lastly, the CMC module combines spatial and temporal information into a unified representation which is used to perform sports events categorization based on athlete movements. The proposed model is evaluated on Olympic sports dataset and University of Central Florida’s (UCF) Sports dataset across distinct evaluation measures. The model achieved 98.36% accuracy, 99.42% precision, 98.42% recall, 98.91% F1-score on the Olympic Sports dataset and 98.64% accuracy, 98.45% precision, 98.91% recall, and 98.68% F1-score on the UCF Sports dataset. Moreover, this method showed supreme results compared with existing techniques in all metrics, demonstrating its effectiveness and potential for high- applications in sports analytics and athlete monitoring.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring the potential of deep learning techniques for analyzing athlete movements in competitive athletics sports

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Yilun Gao,
  • Jie Zou,
  • Yuexin Zhang

摘要

In this research work, a Deep Learning (DL) approach utilizing Spatial Transformer, Temporal Transformer, and Collaborative Movement Centric (CMC) module is presented to classify sports event based on athlete movements in competitive sports. Primarily, the Spatial Transformer utilizing the Spatial Feature extraction module (SFEM) is introduced to extract detailed spatial information from video frames. Here, the SFEM employs Modulated Moving Average Graph Convolutional Network (MMA-GCN) to extract complex spatial relationships by learning the knowledge from offset and modulation parameters. Secondly, the Temporal Transformer is designed that employs Temporal Feature extraction module (TFEM) module to extract long-range temporal dependencies and model evolution across consecutive frames. Lastly, the CMC module combines spatial and temporal information into a unified representation which is used to perform sports events categorization based on athlete movements. The proposed model is evaluated on Olympic sports dataset and University of Central Florida’s (UCF) Sports dataset across distinct evaluation measures. The model achieved 98.36% accuracy, 99.42% precision, 98.42% recall, 98.91% F1-score on the Olympic Sports dataset and 98.64% accuracy, 98.45% precision, 98.91% recall, and 98.68% F1-score on the UCF Sports dataset. Moreover, this method showed supreme results compared with existing techniques in all metrics, demonstrating its effectiveness and potential for high- applications in sports analytics and athlete monitoring.