Scholarlink

<p>In table tennis training, pose-based motion analysis is of great significance for technical evaluation and training feedback. With the development of Artificial Intelligence (AI), pose estimation provides a new technical approach for real-time and refined motion analysis. This study proposes a Lightweight Attention-enhanced Fusion Pose Estimation Network (LAFPose), which is improved based on YOLOv8m-Pose. The model adopts MobileNetV3 as the backbone feature extraction network, introduces the Convolutional Block Attention Module (CBAM) and the adaptive key point enhancement module, and replaces the up-sampling module with the Content-Aware ReAssembly of Features (CARAFE) module. These designs make the network structure more lightweight and enhance its feature expression capability. Experiments on table tennis videos from the University of Central Florida 101 (UCF101) dataset show that LAFPose achieves an accuracy of 86.8% with a model size of only 33.2&#xa0;MB and a computational cost of 46 GFLOPS, achieving a better balance between lightweight performance and precision. In the empirical study, 120 athletes receive AI system intervention. Three groups are designed: the real AI intervention group, the false feedback control group, and the traditional training group. The results show that the total motivation score of the real AI intervention group increases from 18.45 to 20.75, and its satisfaction score rises from 3.62 to 4.21. Both scores are significantly higher than those of the other groups (<i>p</i> &lt; 0.001). Cohen’s d reaches a large effect size. The results show that the pose-driven motion analysis and real-time feedback mechanism supported by LAFPose exhibit excellent performance in computational efficiency and analysis accuracy, and significantly enhance athletes’ participation motivation and training experience. It holds important practical value for the design of intelligent sports training systems and sports psychology research.</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

Motion analysis driven by table tennis pose and analysis of participation motivation and athlete satisfaction based on artificial intelligence YOLOv8

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月15日

查看全文
  • Kaihao Yu,
  • Shamsulariffin Bin Samsudin,
  • Mohd Aswad Ramlan,
  • Faizal Bin Abd Manaf,
  • Yuxin Cong

摘要

In table tennis training, pose-based motion analysis is of great significance for technical evaluation and training feedback. With the development of Artificial Intelligence (AI), pose estimation provides a new technical approach for real-time and refined motion analysis. This study proposes a Lightweight Attention-enhanced Fusion Pose Estimation Network (LAFPose), which is improved based on YOLOv8m-Pose. The model adopts MobileNetV3 as the backbone feature extraction network, introduces the Convolutional Block Attention Module (CBAM) and the adaptive key point enhancement module, and replaces the up-sampling module with the Content-Aware ReAssembly of Features (CARAFE) module. These designs make the network structure more lightweight and enhance its feature expression capability. Experiments on table tennis videos from the University of Central Florida 101 (UCF101) dataset show that LAFPose achieves an accuracy of 86.8% with a model size of only 33.2 MB and a computational cost of 46 GFLOPS, achieving a better balance between lightweight performance and precision. In the empirical study, 120 athletes receive AI system intervention. Three groups are designed: the real AI intervention group, the false feedback control group, and the traditional training group. The results show that the total motivation score of the real AI intervention group increases from 18.45 to 20.75, and its satisfaction score rises from 3.62 to 4.21. Both scores are significantly higher than those of the other groups (p < 0.001). Cohen’s d reaches a large effect size. The results show that the pose-driven motion analysis and real-time feedback mechanism supported by LAFPose exhibit excellent performance in computational efficiency and analysis accuracy, and significantly enhance athletes’ participation motivation and training experience. It holds important practical value for the design of intelligent sports training systems and sports psychology research.