网球比赛视频分析的若干技术研究

【作者】 董敏；

【导师】 王建宇；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 随着视频数据的飞速膨胀,如何找到所需的信息,已经成为一个急需解决的问题。本文面向体育视频领域,研究了基于内容的视频分析和检索技术。其目的就是通过对视频内容进行计算机处理、分析和理解,建立结构和索引以方便用户获取,具有重要的理论意义和应用前景。本文以网球比赛视频为对象,主要对下面几个问题进行了讨论,包括镜头边界检测、关键帧选取、特征提取,以及镜头分类、音频类型识别、网球视频感兴趣事件的检测。镜头边界检测是进行视频处理的第一步,本文在总结了现有镜头边界检测方法的基础上,采用了一种基于动态阈值的镜头突变检测算法,实现了对突变镜头的检测;采用一种基于灰度级平均的方法,实现了对渐变镜头的检测。在对每个镜头选取关键帧的基础上,本文提出了一种基于K均值聚类结合关键帧主色率统计的镜头分类算法,完成了对比赛镜头的筛选,并在此基础上,采用一种基于场地颜色的场地类型检测方法,实现了网球比赛场地类型的检测。对于镜头分类得到的比赛镜头,提取出音频流,采用SVM决策树的方法,实现镜头中有意义的音频段的识别。然后,综合音视域的分析结果,实现网球比赛精彩事件的检测。最后,本文以Visual C++6.0和Matlab 7.0为开发平台,实现了一个网球视频自动分析原型系统。实验表明,本文采用的网球视频语义分析算法具有令人满意的效果。更多还原

【Abstract】 With the rapid increase of the amount of digital videos, the problem how to find useful information has become much urgent. This paper focuses on content-based video analysis and retrieval technology in sports video domain. The aim is to process, analyze and understand video content with computer to construct index and structure for facilitating user’s access, which is provided with important academic appeals and commercial potentials.This paper takes tennis video as research object, and discusses several problems in the process of content-based video retrieval, including shot boundary detection, key frame selection, feature extraction and shot classification, audio type classification, interesting events detecting.Shot boundary detection is the first step of video processing, starting from a generalization of methods of shot boundary detection, this paper analysised an algorithm for detecting abrupt video shot boundaries based on method of adaptive threshold, and an algorithm for gradual shot boundaries detecting using the mean gray level (MGL) of image.Based on abstracting typical frames of video shots, this paper proposed a shot classification method which is based on K-means cluster processing combined with main color rate computing. After that, this paper used a algorithm for detecting court class based on court color.For play shots, important auditory features including both ball hitting and cheer are detected by using SVM. Afterwards, audio features are applied into play shots for interesting events detecting in tennis video.Finally, This paper implements a prototype system for content-based tennis video anlysis by Visual C++ 6.0 and Matlab 7.0, and the experiments have demonstrate that all these methods are effective.更多还原