Object Tracking Based On Structured Sparse Representation And Collaborative Model

Target tracking has widely used in military guidance, intelligent transportation, videocoding, human-computer interaction and so on. Target tracking is a necessaryand fundamentaltask of intelligent video analysis and provides a reliable basis for high-level visual analysis.Tracking algorithms have important economical and theoretical value. In this paper, wesummarize the tracking algorithms and carry out an in-depth research on target trackingalgorithms based on sparse representation. Moreover on the basis of summing up the previousworks, we put forward our algorithm called object tracking based on structural sparserepresentation and collaborative model for challenge issues of target tracking.For the purpose of being robust to appearance change and exploiting both local and spatialinformation of the target, structure sparse representation is introduced to code the imagepatches in the target. Then in the discriminative model we multiple the reconstruction residualof sparse coding and score of the Naive Bayes model value to get the confidence value, thismethod can effectivelydistinguish between foreground and background with different weights.In generative model, we use averaging, anti-occlusion handling, alignment pooling andweighting operations on obtained sparse coding coefficients to preserve spatial and localinformation of the target to improve tracking accuracy.we make full use of the advantages ofboth discriminative and generative models to achieve a more reliable likelihood function in thecollaborative model. In addition, we use a dictionary update strategy based on incrementalsubspace learning and sparse representation for effectively alleviating the impact of driftproblem.Wetimelyupdatecovariancematrix ofthejoint Gaussiandistribution accordingto theprevious tracking results to achieve an adaptive motion model for target state analysis. Finally,combination of appearance model based on structure sparse representation and affine motionmodel forms the particle filter framework to achieve robust object tracking.Experiments on some challenge video sequences demonstrate that our proposed tracker isrobust and effective to challenge issues such as illumination change, clutter background,partial occlusion and so on and perform favorably against state-of-art algorithms.