足球视频镜头分类方法  

Method of shot classification for soccer video

作　　者：林彬[1] 刘群[1] 王群[1] 聂燕柳[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《计算机工程与设计》2012年第4期1467-1471,共5页Computer Engineering and Design

基　　金：重庆市自然科学基金项目(CST2007BB2386);重庆市科委应用基础研究基金项目(KJ070504);重庆邮电大学博士启动基金项目(A2009-10)

摘　　要：根据视频语义分析和视频摘要等应用对于视频数据结构化的需求,提出了一种针对足球视频的镜头分类方法。通过logo模板匹配检测并定位出视频中的慢镜头,对其余的正常比赛部分做镜头边界检测完成视频切分。基于分块的思想,对正常比赛镜头帧计算其各块的场地像素比率值作为特征,利用SVM分类器将正常比赛镜头分为远镜头、中镜头、球员特写或场外镜头3类。至此,整个视频流可以表示为结构化的四类镜头类型标示序列。实验结果表明,该方法在视频切分和镜头类型识别的准确性方面具有良好的效果。A novel method is proposed for soccer video shot classification. Firstly, slow-motion replay segments detection is con- ducted by determining the logo template based on user interaction and by locating all the similar frames in the video using the logo template. For the remaining part of the normal game, detect abrupt transition to complete the video segmentation. Then, based on the SVM classifiers, the normal game shots are classified into three classes： long shot, medium shot and close^up or out of field shot. Thus, the video stream could be expressed as a structured sequence of four types of shots. Experimentation shows that the approach performs good performance in the accuracy of video segmentation and shot classification.

关 键 词：足球视频 镜头分类 视频切分 慢镜头 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]