摄相机运动识别及视频分割的组合检测新方法

针对视频镜头分割处理的需求，提出了一种综合运动参数曲线法．它以由压缩域视频流中运动信息估计出的摄像机运动参数为基础，对视频流进行基于内容的分析．该方法的特点是在对视频流处理中将摄像机的运动识别与视频分割问题综合考虑，在求取帧间摄像机运动参数以及判断连续帧间的摄像机运动趋势的同时，可一并实现对视频流时域上的自动分割，以及近景镜头等的识别，对各种类型视频均采用相同的技术操作和统一的门限值，具有上佳的鲁棒性，对MPEG组织专用于内容检索的10类共计284min的标准视频进行了实验，查全率达92％～100％，查准率也高达86％以上(动画及体育片段除外)，证实了所提方法的有效性。

New Method of Combining Camera Motion Recognition and Shot Change Detection

An integrated motion-parameter-curve algorithm is presented for content-based video analysis, which implements shot segmentation based on camera motion parameter estimation with motion information of compressed domain. The algorithm facilitates calculating camera motion parameters between frames, and judging camera motion trends within consecutive frames; realizes automatic temporal segmentation and recognizes close-shots in video streams simultaneously. The advantage of this algorithm lies in the integrated consideration on camera motion recognition and shot change detection, adopting the same technique and the uniform threshold sets for various video types robustly. Ten paragraphs of standard video sequences of 284 minute long are tested, and the recall rate reaches from 92% to 100%, the precision rate is higher than 86%-not counting cartoons and sports snatches.