| For our human, it is the main way that we get information by our eyes. In recentdecades, Image processing technology has been rapid development. As one of the keytechnology in the field of image processing, object tracking has been widely used in thefield of civil and military and so on. So we are focused on the study of algorithm formoving target tracking and it has important theoretical significance and practical value.This article is based on those demestic and foreign outstanding achievements, studydeeply on the characteristics of image, and then put forward some improved algorithms inthe framework of Mean Shift algorithm.Firstly, The traditional Mean shift algorithm is implemented by the C++programming language. Then experiments prove that this code is right and valid. So thecode can be used in practical application.secondly, In view of the Mean Shift algorithm itself cannot be used to the targetobject which can dynamically change its size and direction, and other algorithms eitherthe calculation is complicated or the result is bad. In this chapter, weighted-backgroundhistogram has been used to make the weights more robust and close to the actualdistribution of the image. Image moment theory is also introduced in, and it is used toestimate the change of the target on size and direction. All these work will have asignificant impact with Mean Shift algorithm, and thus it is more robust and less timeconsuming.Thirdly, In view of the previous Mean Shift algorithm failed to make full use ofimage’s texture feature and structure information, This chapter describes an alternativehybrid scheme, globally rotation invariant matching with locally variant local binarypattern (LBP) texture features. A new texture descriptor, LBP variance (LBPV), isproposed to characterize the local contrast information(ie the global structural informationin target area) into the one dimensional LBP histogram. Compared with other algorithms,there have more robust and less time in this algorithm. |
PDF Full Text Request