Research On Target Tracking About Moving Human And Behavior Recognition In Video Sequences

Intelligent video surveillance is able to achieve intelligent "monitoring" onunattended surveillance scenes with the video analysis technologies. It will alarm assoon as the defined anomalous behavior appears in the surveillance scenes, whichovercomes the main problem of the traditional monitoring system---that traditionalsurveillance system always can only deal with the problem after it has happened. Thedetection and identifying of moving target is an important study aspect in theintelligent surveillance research. This paper mainly focuses on detection andrecognition of the moving body target in the surveillance scene monitored by the fixedmonocular camera.Firstly, this paper will distinguish the foreground region and background regionwith methods of traditional frames subtraction and background subtraction, and thenuse the method of statistical average to update the background, which guarantees theaccurate detection for moving targets. As for the problem that moving targets oftencontain the shadow, this paper uses the principle of human vision to distinguish thetarget and the shadow produced by target so that we can suppress the shadowsuccessfully because of the difference between intensity and hue of HIS color space.Secondly, after the research on traditional mean-shift algorism, this paperproposes the mean-shift algorism based on edge detection of moving targets to updatethe width for the kernel window, which overcomes the defect that the width for thekernel window can not be changed with the changing size of the moving target. Thenit also proposes the real-time detection and tracking of moving targets, which meanstracking the moving target instantly enables the tracking. The real-time detection andtracking for moving targets is better by the comparison and analysis of the twomethods. However, it is easy to miss the tracking targets when the target stopssuddenly.Finally, this paper introduces the method of judging the similarity betweenimages with the technology of regulations. Then this paper focus on the analysis onsilhouette features and the centroid for different behaviors, and puts forward a methodwhich converts the silhouette features into one-dimensional features by calculating.Finally, we propose a behavior recognition method which combines the imagesimilarity judgment and silhouette features similarity analysis. By several simpleexperiments for recognition the human behaviors it shows that: this method has a better recognition result for some behaviors with big differences, and the recognitionrate will gradually increase with the increase of frames number, the recognition ratewill be stabilized after the frames in video reach50.