Research On Vedio Moving Target Tracking Based On Kalman Filter

Computer vision is a challenging and important research problem in both engineeringand science field, and that the moving target in vedio is exactly the important part ofcomputer vision. Vedio moving target tracking combine military, public and other fields ofanvanced technology, includes image processing, artificial intelligence, pattern recognition,automatic control, medical treatment diagnosis and so on. It has been extensive applied invisual guidance in the military, robot vision navigation, safety inspection, public scenemonitor, intelligent transportation and other fields. This makes many researchers have beenconcentrating on the technical research from home and abroad. But because of the influencefrom many factors such as light change, noise, shelter, same color and so on, the existingalgorithms face many problems in practical application. Therefore, research and designdependable, global moving target tracking algorithm still face a tremendous challenge.This thesis has researched the moving target detecting methods base on the analysis ofpresent development of the target tracking technology, contrasted several commonly useddetecting algorithms of moving target include interframe difference, background difference,optical flow, and intensive studied the three frame difference by interframe difference. Putforward a modified automatic moving target detecting and dividing method by research, theexperiment results showed that the method could effectively detect moving target, providetarget source for tracking system.The basic technology of moving target tracking has been researched in this thesis,include classical MeanShift algorithm, CamShift algorithm, then derivate the algorithmsand analysis their relative merits. Through the simulate experiment, CamShift algorithmwhich based on the MeanShift algorithm has preferable tracking result. At the same time,the kalman filter and extended kalman filter has been researched for analyze how theyachieve target reckoning and their autoregressive data processing method, and apply inadvance and improve the accuracy of tradition target tracking algorithm.Through the research on CamShift moving target tracking algorithm, the target willlose when the target encounters seriously shelter or meets large area of similar colorinterference. Contrapose the drawback of this algorithm, a modified CamShift algorithmhas been put forward which base on the self-adaptive kalman filter for the purpose ofresolve target tracking failure problem. The method above acquires position and colorinformation of the target from detecting and dividing by the three frame difference algorithm, then kalman fileter was used to estimate the initial iteration position of everyframe according to the detecting position. Finally, get the convergent iteration positioncalculate by CamShift algorithm, and transfer the iteration result to kalman filter for thenext position estimation. When the target encounters seriously shelter or meets large area ofsimilar color interference, in this thesis, put forward self-adaptive factor to dynamic adjustthe parameter of kalman filter according to the interference degree, consequently make themodified algorithm also have the capability to calculate the succeed state of the movingtarget. The experiment result demonstrated that the algorithm has great improvementrelative to the traditional target tracking algorithm, when the target encounters seriouslyshelter or meets large area of similar color interference, the algorithm can also exactly andcontinuously track the moving target, improve the accuracy and robustness of targettracking algorithm.