The Research Of Shot Segmentation And Target Tracking In Video Analysis

Video surveillance is an important subject in the computer vision sector. The wide spread of computer ability and the gradual materiality of computer vision algorithm, along with the camera and the growing developed functions, make the sector more developed.The huge information volume of the video makes it hard to monitor. Besides, human surveillance is of great expense. As a result, we have to develop effective algorithm to make the computer more intelligent in monitoring the video.The first step of the video surveillance is to analyze the content of the video and extract effective information from video frame as the video is made up of frame pictures.Frame pictures are distinguished by color, texture, fringe and motion, which we can use to analyze the content of the video through shot segment, key frames and image character extracting, target tracing and behavioral analysis.This thesis primarily deals with shot segmention and target tracing. It gives brief introduction to the extract of key frames and experimental comparison among several ways of key frames extraction. In addition, a simple description of image feature extraction and the combination use of several features are also included.In terms of shot segmention, this thesis still bases on the color histogram and includes edge information, which not only calculated the difference of the color histogram, but also compared the change rate of the edge point in comparing the difference between two frames image. This makes shot segmention more accurate and reliable.In respect of target tracing, this thesis still bases on Mean-Shift, but also provide a new color histogram model, which takes full consideration of the edge of pixel and positional information. Consequently, the color histogram will contain more information and yield better effect.