基于机器视觉的乒乓球正反手动作分类及计数研究  

Research on the Classification and Counting of Table Tennis Forehand and Backhand Movements Based on Machine Vision

作　　者：焦禹铭 李荣芝 JIAO Yu-ming;LI Rong-zhi(Shanghai University of Sport,China Table Tennis College,Shanghai,200438)

机构地区：[1]上海体育大学,中国乒乓球学院,上海200438

出　　处：《吉林体育学院学报》2025年第1期100-108,共9页Journal of Jilin Sport University

基　　金：上海市哲学社会科学规划专项课题(项目编号:2023VZH030)。

摘　　要：在竞技体育智能化发展趋势下,乒乓球运动员技术动作的精细化分析成为提升训练效能的关键环节。针对传统基于人工观察的动作分析方法存在的效率低、主观性强等局限性,本研究提出一种基于改进YOWO框架的乒乓球动作识别模型。采用Darknet-53增强模型对视频帧空间信息的提取能力,并引入SKNet注意力模块以增强模型对小目标的检测能力,使用CBAM注意力机制解决YOWO模型在特征融合部分对两个分支的特征信息融合不充分的问题,拟改进乒乓球基础动作分类模型。结果显示:(1)消融实验,改进的YOWO模型动作分类准确性高于原始YOWO模型和部分改进的YOWO模型;(2)对比实验,改进的YOWO模型动作分类准确性高于I3D、C3D和原始YOWO模型。结论:对YOWO模型2D CNN分支和特征融合的改进,可有效提高模型在乒乓球基础动作分类的准确性。With the intelligentization of competitive sports,the refined analysis of table tennis players'technical movements has become critical for enhancing training efficiency.To address the limitations of traditional manual observation-based motion analysis methods,such as low efficiency and high subjectivity,this study proposes an improved YOWO-based table tennis action recognition model.The model employs Darknet-53 to enhance the spatial feature extraction capability of video frames,introduces the SKNet attention module to improve the detection of small targets,and utilizes the CBAM attention mechanism to address the insufficient feature fusion of the two branches in the YOWO model,aiming to optimize the classification of basic table tennis actions.The results show that:(1)In ablation experiments,the improved YOWO model achieves higher action classification accuracy than the original YOWO model and partially improved YOWO models;(2)In comparative experiments,the improved YOWO model outperforms I3D,C3D,and the original YOWO model in action classification accuracy.The conclusion indicates that the improvements of the 2D CNN branch and feature fusion of the YOWO model effectively enhance the accuracy of the model in the classification of basic actions in tabletennis.

关 键 词：乒乓球运动 动作识别 深度学习 YOWO模型 

分 类 号：G846[文化科学—体育训练]