Online Visual Saliency Map Based Tracking Method Via Structural Inverse Sparse Appearance Model

With the development of computer vision and image processing,visual tracking has been widely applied in many fields such as security surveillance,intelligent transportation,vision navigation,robot control and so on.Although researchers have proposed many different types of tracking algorithms,achieved poor robust and real time tracking results on challenging video sequences with occlusion,illumination changes and the interference of background which result in the changeable of the target appearance,it is remain a challenging task to achieve a robust and real time tracker.This paper focus on how to improving the performance of Meanshift algorithm,particle filter and sparse representation based trackers.The main research problems of this paper are as follows:First of all,considering that the traditional Meanshift utilizes center-weighted color histogram as the reference model which is susceptible to interference of background pixels,and that can result in the compromised tracking robustness,we propose a midlevel cues mean shift visual tracking algorithm based on target-background confidence map saliency-weighted model.A discriminative appearance model based on superpixels is introduced,thereby it can facilitate a tracker to distinguish between target and background by different weights and it improved the performance of the tracker.Secondly,in order to improve the performance of sparse representation based trackers,an object tracking method based on inverse sparse representation with combining the global and local information is proposed which develops the tracking framework based on a Bayesian and particle filter framework.And this method utilizes an inverse sparse representation formulation which enables the tracker to compute the weights of all particles by solving one optimization problem and this is conducive to improving the real time performance.We designed a ranking mechanism based on joint discriminative similarity map(JDS map)to measure the similarity of candidates and the template,and which scores highest is the tracking result.In the procedure,the template and weight template are updated using a new double-template updating strategy.The proposed method is robust when targets in the complicated conditions.Finally,an online visual saliency map based tracking method via structural inverse sparse appearance model is proposed which develops the tracking framework based on a Bayesian framework.The inverse sparse representation with combining the global and local information utilizes a correlation visual saliency detection method based on Markov model to calculate the saliency map of the target template of the current frame and obtains the adaptive weight of each pitch by mapping the saliency map to the candidate pitches.Finally,a novel combine mechanism is utilized to unite the global and local sparse solutions which is applied to measuring the similarity between the candidates and the template,then the optimum target state can be estimated and tracked under the Bayesian framework.