The Prospect Of Moving Target Extraction And Tracking In Video Image

The prospect of moving targets extraction method and moving target tracking algorithm invideo image has always been an important research content in the field of computer vision. In theintelligent transportation monitoring, video security monitoring and navigation, it all has importantpractical value.In intelligent video monitoring processing, foreground extraction is undoubtedly a key link ofinformation extraction. Its task is to extract all the real motion in the video target and eliminatevarious noise interference and the false targets due to other reasons. It is obvious that this step willbe the premise of subsequent target recognition tracking and behavior analysis. The third chapterbriefly introduces the prospect of commonly used three kinds of extraction methods: optical flowmethod of frame differential method and background subtraction division. Combined with thesealgorithm’s advantages and disadvantages then proposed a new algorithm named dynamic numberof gaussian mixture background modeling algorithm. Traditional gaussian mixture model iscommonly used in background subtraction division of background modeling algorithm, which takeon the number of gaussian distribution, generally for a fixed value between3~5. But in thepractical application scene of digital image processing, a fixed value of gaussian mixture modeloften effects not beautiful. This paper propose a dynamic number of gaussian mixture backgroundmodeling algorithm which can change based on real-time scene image area. Spontaneousdynamically adjust the number of gaussian distribution, so as to improve the effect of backgroundextraction and the efficiency of algorithm.After video image foreground extraction success, tracked targets have been identified in thecontinuous frames in video sequences. So the fourth chapter began to introduce the traditionalMean-shift target tracking algorithm based on kernel function principle. And track extract of theimproved background modeling algorithm prospect in the third chapter based on Mean-shifttracking algorithm. The experimental results shows that the traditional Mean-shift trackingalgorithm’s effect is well when wide window in tracked target size not rigid size changes. However,for some of the more complex scenarios, such as the increasing size of the rigid target tracking, it iseasy to cause the failure of tracking. So chapter v of this article put forward a kind of relativelynovel nuclear window width automatic selection algorithm. To cope with the scenario that thetracked target is increasing the size. The method USES affine model, after tracking-centroidregistration and feature points matching and return to realize the adaptive bandwidth Mean shifttracking algorithm. Successful track the increasing size of the rigid target. Experimental resultsimproved significantly than traditional Mean shift algorithm.