Research On Video Surveillance Based Special Object Tracking Algorithm

To improve intelligent information level of the security---especially for the special security area,it is very crucial to implement the automatic visual tracking for the security safety.In order to improve the public management of efficiency,this paper presents some researches on special target visual tracking under the real-time surveillance.The research structure of this paper contains:(1)Research on real time visual tracking algorithms,especially for the target object appearance model construction and matching scheme representation.(2)Research on basic theory about sparse representation and feature fusion:for the low-accurate appearance model description problems in traditional visual tracking algorithms---illumination changes,shape changes,occlusion,motion appearance variation,this paper describes the object appearance with compact features which could improve the compactness of object inner appearances and accuracy of the model.Meanwhile,local and global coding schemes are constructed by multi-features in the sparse modeling process,and different constraints are applied for sparse decoding process in order to track the target object efficiently.In detail,co-variance matrix and Log Polar Histogram of Sift descriptor are coded in local and global level for the appearance of target region of interest.The compactness structure of features dig out the latent attributes in the region of interesting,and then local and global model can be explained by the LASSO and SOMP algorithm,separately.All those solutions give a guarantee about the strong relationship between local and global level in the appearance model.(3)Finally,In order to show the efficiency and accuracy of the algorithm this paper gives some simulations about the algorithm based on software and hardware which is depended by the Matlab2017b and MexOpencv-lib that both of them own very huge computer vision functions.