Adaptive Sparse Tracking Based On Particle Filter

Visual tracking is an important branch of computer vision field.It has been widely ap-plied in video surveillance,robotics,human-computer interaction,traffic track and so on.However,tracking an object steadily and precisely remains to be a very challeng-ing task because of heavy corruptions such as deformation,rotations,occlusions and illumination changes.In this paper,we propose an adaptively weighted sparse tracking approach based on particle filter to improve the accuracy and robustness of the tracker.At the same time,we use APG(Accelerated Proximal Gradient)algorithm to reduce the computational expense.Sparse representation have been applied successfully to visual tracking because it is insensitive to a wide array of image corruptions,especially partial occlusions.These tracking methods aim to sparsely represent the tracking candidates with templates.They search for the region with highest likelihood by solving l1 mini-mization problems and calculating reconstruction error.However,traditional methods treat all templates in the dictionary equally regardless of their resemblance to the tar-get.In fact,templates with higher similarity to the tracked objected should be weighted more.Thus,an adaptively weighted appearance model can represent the target more ac-curately.Another significant part of a tracking system is motion estimation between frames.The particle filter is a Bayesian sequential importance sampling technique for estimating the posterior distribution of state variables characterizing a dynamic system.It can estimate and propagate the posterior probability of the target without knowing the concrete observation probability and for this reason it has been introduced to vi-sual tracking system.The superiority of our system over current state-of-art tracking methods is demonstrated by a set of comprehensive experiments on public data sets.