Research Of The Moving Target Detection Algorithm Based On Global Motion Compensation

The detection of moving object is the basis of intelligent visual surveillance andthe analysis of target behavior, and there are broad application prospects in the fieldsof military, transportation, industrial, and biomedical. The main purpose is to extractthe moving target area accurately from video images. However, in practicalapplications, with the background changes caused by camera motion, camera shake,and the irregular changes of the outside environment, these factors bring the greatdifficulty to detect the moving target detection from the video. In this paper,in orderto solve the above problems, through inquiry learning detection of moving target,global motion estimation and compensation, and digital image processing, and finallyon the subject that the moving target detection method which based on global motioncompensation to do more in-depth study, and the main study results and innovationsare as follows:Firstly, to prevent inaccurate motion vectors and avoid trapping in local optimum,a method of block matching motion estimation based on genetic algorithm is proposedin this paper. Experiment results show that the method has greater precision comparedwith some other fast block matching algorithms.Secondly, In order to provide available credible data for building an accuratecameral parameter system, a global motion estimation algorithm based on thereliability analysis of motion vectors is proposed in this paper.Thirdly, building a cameral parameter system using reliable motion vectors, theglobal motion estimation parameters are obtained by resolving the over-determinedlinear system equation using iterative linear least-square method. Experiment resultsshow that the proposed method can extract accurately the global motion estimationparameters, and compared with the tradition method, our method improves the speednearly one times, while the accuracy is guaranteed.Finally, In the process of detecting moving object segmentation, In order toreduce the over-segmentation of the automatic threshold segmentation algorithm, amethod combined with image enhancement about automatic threshold segmentation isproposed.Experiment results show that this paper’s method can effectively extract movingobjects from video sequences with moving background. It can provide fundamentalelements and analytical data for application of video motion information analysis, andit also fits the smart surveillance with Pan/Tilt camera.