Study On Target Tracking Based On Compressive Sensing

Target tracking is always the key issues that need to be solved in the video security field.Many scholars have deep research on target tracking algorithm,and also have some achievement.Target tracking algorithm based on compression perception is put forward in recent years,a fast tracking algorithm,Measurement matrix is generally selected for dimension reduction goals.The selection of measure matrix's sparseness and target's search range is fixed commonly in compressive sensing algorithm,it can't be change d along with the size of target,and affect results eventually.Those tracking problems,this paper such as the construction of measurement matrix,the selection of sparse degrees,the affect of measurement dimension number,the construction of classification device,features fusion amendment and track window since adapted search range and so on,and using that to solve track failed problem in the process of target track,improving algorithm of robustness eventual.(1)This algorithm is discussed about by choosing a different measurement of the sparseness of the matrix to find the best,to a certain extent,random measurement matrix is a certain degree of adapt ability.Then under the condition of the same measurement matrix is sparse,discusses the compression in the process of matrix measurement dimension for tracking effect.(2)On this basis compression tracking algorithm proposing a new algorithm based on random measurement matrix compression tracking improvement.This algorithm first prediction about the target position,when the classifier to build,update learning parameters setting the threshold value,classifier for classification,the classification score maximum sampling location as the target track position.Classifier to update the original algorithm learning parameters constant,easy to cause the loss goals late tracking process.(3)Proposing Based via speed feature in fast compression tracking.Most compression algorithm extracts only describe the Haar-like characteristics of the target as a discrimination of target,the effect of trace is lack of robustness.In addition,the compression algorithm searches targets according to a fixed size of search box,it is easy to overstep the range of computation when meeting the targets are moving fast,eventually causing the track to fail.In order to solve thisproblem,this article proposes fusion target velocity correction method and combines with particle filtering framework hasn't been done,providing the robustness of the algorithm,it also uses the s-shaped curve in final classification,the classifier parameters uses the way of updating adaptively,obtaining the best target tracking results.