Research On Object Tracking Algorithm Based On Sparse Representation And Feature Selection

Object tracking is an important research content in t he field of computervision,which has also become a research hotspot in recent years.It has many important applications such as video surveillance systems,intelligent traffic monitoring,human-computer interaction and military.This paper proposes a object tracking algorithm based on local sparse representation,mainly aiming at the existing problem of tracking a moving target.It is According to the characteristics of the local sparse representation,and using local sparse representation to the target mode l.The particle filter framework is combined with the Bias classifier to track the target for a long time.The tracking is achieved under different conditions such as occlusion,illumination change,fast moving target and background similarity.In addition,the algorithm also uses a shade interference robustness update strategy which can be removed and good results have been achieved.The main work of this thesis is as follows:?On the basis of analyzing and summarizing the research status of moving object tracking,the particle filter theory and sparse representation theory are summarized.In this paper,we introduce the theory of sparse representation and its application.To commonly used in object tracking method at the same time from the principle and experimental results are given in detail.?Combining the advantages of both the generative tracking algorithm and the discriminative tracking algorithm,a new algorithm for object tracking based on local sparsity is proposed.The local sparse coefficient is used as the feature representation of the target,and the two step search strategy is used to locate the position of the target accurately using the naive Bias classifier.In the model updating,using the update strategy of anti occlusion,i.e.in the presence of occlusion is the use of smaller weight information of the current frame is blocked,and no larger weights using the current frame information block,this method has good adap tability to occlusion.Through different algorithm is compared with the proposed algorithm,the experiment of justifying the proposed algorithm has a good tracking effect and other issues in the face of occlusions,illumination change,fast motion and other issues.? Put forward the local sparse tracking algorithm based on feature selection,further improved based on local sparse tracking algorithm.In this algorithm,the feature selection process is added on the basis of local sparse tracking algorithm.It is composed of positive negative class variance ratio and mean difference.In the paper,we propose a method to solve the problem of fixing the threshold by using the differenc e method,which is a positive response to the stability of the tracking.