体育视频中动作识别技术研究综述  

A review of research on action recognition methods in sports video

作　　者：游义平 季云峰[2] YOU Yiping;JI Yunfeng(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Institute of Machine Intelligence,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093 [2]上海理工大学机器智能研究院,上海200093

出　　处：《智能计算机与应用》2023年第11期1-13,共13页Intelligent Computer and Applications

基　　金：国家自然科学基金(61773083);上海市浦江人才计划(2019PJD035);上海市人工智能创新发展专项;上海市引进海外高层次人才工作专项;上海高校特聘教授(东方学者)计划。

摘　　要：随着中国成功举办多项国际体育赛事以及互联网短视频平台的兴起,视频数据呈爆炸式增长,且体育运动越来越受到人们的关注,体育视频中的动作识别成为计算机视觉研究的一大热点问题。本文综述了体育视频中动作识别技术现有应用与研究方法,第一部分回顾了近年来动作识别在体育赛事中的应用现状,将其归纳为辅助判罚、精彩动作集锦、体育新闻自动生成。第二部分总结了体育视频动作识别相关数据集。第三部分回顾了近年来动作识别在体育视频中的实现方法,将其总结为基于传统手工特征的算法和基于深度学习的算法,基于深度学习的算法将其归纳为基于2D模型、基于3D模型、基于双流/多流模型、基于Transformer模型,并总结了各模型的优缺点。最后,讨论了体育视频动作识别的难点与挑战。With the successful hosting of many international sports events in China and the emergence of Internet short video platforms,video data is exploding and sports are getting more and more attention.Action recognition in sports video has become a hot topic in computer vision research.This paper reviews the existing applications and research methods of action recognition technology in sports video,and the first part of this paper reviews the current situation of action recognition applications in sports events in recent years,and summarizes them as auxiliary penalty,highlight action collection,and automatic sports news generation.The second part summarizes the data sets related to sports video action recognition.The third part reviews the implementation methods of action recognition in sports video in recent years,summarizes them as traditional manual feature-based algorithms and deep learning-based algorithms,and the deep learning-based algorithms are categorized as 2D model-based,3D model-based,two-stream/multi-stream model-based,and Transformer model-based,and summarizes the advantages and disadvantages of each model.The final part discusses the difficulties and challenges of sports video action recognition.

关 键 词：动作识别 深度学习 体育运动 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]