一种基于运动的视频层次化分析框架

A Video Hierarchy Analysis Framework Based on Motion

运动信息是视频中最重要的特征之一.本文提出了一种新颖的基于运动的视频层次化分析框架.为了揭示视频流中的运动关系和在拍摄过程摄像机的运动信息即拍摄者的兴趣所在,提出了视频层次化结构来对视频流进行描述和检索.该运动层次化结构引入一个新的概念:基本运动单元(Basic Motion Unit,BMU).BMU反映的是MPEG-7中定义的基本摄像机操作,这样就可以把视频分割成BMU的集合.在BMU的基础上发现并挖掘出视频流中潜在的运动语义模式,最后根据运动活力(Motion Activity)将BMU聚类为运动节奏快慢的运动活力单元(Motion Activity Unit,MAU).该基于运动的视频分析框架不仅能够直观地捕获视频流在拍摄过程中拍摄者的注意力,而且为视频流的快速浏览和检索提供了新的思路.

Motion information is one of the most important features in video. A novel video hierarchy analysis framework based on motion is presented in this paper, which can be used to describe and retrieval video stream for reflecting the motion relation and camera motion information during the producing of the video. A new concept is proposed in this framework： Basic Motion Unit （BMU）, which reflect the dominant camera motion type defined in MPEG-7. According to the above idea video can be segmented into the set of BMUs. Then the motion semantic schemes are discovered and mined based on BMU. Finally BMUs are cluster into Motion Activity Unit （MAU） with fast or slow motion rhythm according to the motion activity proposed in MPEG-7. In addition, the presented video analysis framework can not only capture the attention of cameraman but also provide a new approach for the efficient non-linear accessing of video streams.