The Study Of Multi-target Trackingbased On The Union Of Multiple Features

In the past twenty years, with the great progress of information technology, MTT has been widely used in many field s of society and economy. However, there are some common problems for most algorithms, such as the complex background, the occlusion from the shelter, real-time tracking.Mean shift(MS) is a high-speed algorithm of target tracking based on template matching. As convenient and effective, without pre-knowledge, the algorithm is widely studied in target-tracking. However, with the same problem s with other tracking methods, the algorithm is easily interrupted by the similar background, as the result of using the single feature, colour. As well, when the target is obscured by other objects, the algorithm may fail to match the target with too little information.This paper has compared many algorithms of MTT, and studied the advantages and disadvantages of MS al gorithm. Then, the paper has focused on putting forward an improved MS algorithm, which is based on the union of multiple features and with prediction with Kalman filter. At first, we improve the style of sampling feature, and present the feature colour with the background weighted spatio-histogram, and present the feature texture with the local binary pattern(LBP), as well; the paper has proposed an adaptive method to adjust the weight of each feature. Then, we use the Kalman filter to predict the movemen t of the targets, so as to improve the degree of accuracy. And then, with the classification of the event in target tracking, we resize the tracking window and update the model adaptively, and build the whole system of MTT.The paper has checked out the result of series video and camera tracking, and compared with the SVM-MS method. The results turn out that the algorithm in this paper can improve the result of tracking in a big step in complex background, and improve the ability to against the problem of shelter in the degree of 30 percent at least.For the disadvantage of tracking with single camera, with the achievement we have got above, the paper has proposed a scheme of MTT method with distributed multiple cameras. When the target is judged missing, t he scheme will search for it in vision fields of all cameras, continue tracking it with the result of target matching. The result of testing turns out that the method works well.