A Research In Video Based Motion Object Detecting And Tracking

Image sequence based on moving object tracking is a fundamental problem forcomputer vision research and has been widely studied. Visual tracking technique hasmany applications, such as video surveillance, video analysis, video indexing, videobased motion analysis and synthesis, motion-based human identification. Visualtracking technique has made great progress in the past several years, but practicalexperience has shown that visual tracking technologies are currently far from mature.A great number of challenges need to be solved before one can implement a robustvisual tracking system for commercial applications.Under the framework of theory of object detecting based on the backgrounddifference, this paper proposed some useful improved algorithm about backgroundmodeling, threshold segmentation, shadow suppression and object association. Themain contributions of this thesis can be concluded as follows.1: Above the classification of moving object detection algorithm, existingproblem in the tracking progress was analyzed to compare their respective advantageand disadvantage.2: Image segmentation is one of the main content on the moving object tracking.This paper compared the existing performance difference of several segmentationalgorithms and proposed an adaptive threshold segmentation algorithm about theactual object detection, including the good real-time performance.3: Background difference is one of the most general methods of moving objecttracking. Under the actual situation, changing of natural environment will lead to thedisturbance of background and light intensity changes. The paper presents abackground extraction and updating algorithm to solve the problem of backgroundlight changes effectively.4: In order to study the moving object tracking, object tracking algorithm ispresented based on Kalman filter. Through the establishment of inter-frame relationalmatrix, the tracking progress is divided into5cases and analyzed. When occlusionoccurred between two or more object, we are matched by the template ofpre-extraction and obtain accurate objects region. Experimental results show that the algorithm can achieve accurate real-time tracking of moving objects in the situationof complex background.