A Study On Object Tracking Algorithm Based On Sparse Cooperation Model

The task of object tracking is to detect the object area in each frame of a video sequence,and produce a motion trajectory for the tracked object.As an important research branch of computer vision,object tracking has been widely applied in video surveillance,human-computer interaction,and so on.Due to some unfavorable factors,e.g.random sampling,occlusion,appearance variation,etc.,how to achieve a stable and robust object tracking performance is still a relative difficult problem.Aiming at these issues,we have carried out the following two works.Both of these issues are researching single-target in different scenarios.(1)A stable target tracking model is proposed based on a hybrid dictionary.First,two types of dictionaries are extracted respectively by k-means clustering and K-SVD,respectively.Then,the likelihood function values are computed via the sparse collaborative model(SCM)with different dictionaries.In order to decide the final tracking results,a selection strategy is designed by considering three factors,i.e.likelihood function value,reconstruction error,and sparse degree.By fusing the object tracking results obtained via SCM with different dictionaries,the proposed method can effectively alleviate the instability caused by the randomness of sampling.(2)A robust object tracking approach is proposed based on a composite similarity measure.Given the first several frames,an initial dictionary is constructed by searching for the best candidates via the kd-tree function.For a target candidate,a likelihood value is computed by weighting the sparse representation coefficients of local patches obtained via the adaptive structural local sparse appearance(ASLSA)model.In order to mitigate the effects of occlusion,a smoothing operation is designed in computing the likelihood value.On the other hand,on the randomly sampled positive and negative templates,a confidence value is computed via SCM for each candidate.Furthermore,a composite similarity measure is devised by combing the likelihood value and the confidence value.Finally,the best candidate is searched according to the composite similarity measure.The proposed method can significantly improve the robustness of object tracking by combining two similarity measures and a smoothing operation.The stability and robustness of object tracking can be effectively enhanced in our proposed works.The experimental results demonstrated the effectiveness and feasibility of the proposed methods on several widely used and challenging video sequences.