Real-time Robust Object Tracking Based On Sparse Representation And Incremental Weighted PCA

Object tracking is a challenging problem and also an important task within the field of computer vision. It is widely applied for motion analysis, video surveillance, robot navigation and human-computer interaction. In these application scenarios, object tracking is required to be robust and efficient. Although a lot of progress has been made for object tracking in recent years, robust tracking remains difficult in challenging environments. A number of object tracking algorithms have been proposed, but most of them have their strengths and weaknesses, and could not handle all challenging situations.At the beginning the paper reviews the proposed object tracking algorithm at home and abroad, and specific the implementation of these algorithms and analysis advantages and disadvantages of each algorithm. In this paper, we propose a real-time robust object tracking algorithm which is based on a sparse coding tracking framework and weighted incremental PCA model. There are mainly two parts in our research work of this object tracking algorithm.First of all, we propose a target modeling method that is based on a foreground color distribution and weighted incremental PCA model. The foreground color distribution use soft segmentation to solve Gaussian Mixtures Models of the tracking target. The weighted incremental PCA model use Singular Value Decomposition method and foreground color distribution project the appearance of target to low dimension linear subspace. The model can be updated online during tracking process and make it more robust against appearance variations, rotation and illumination change.Secondly, we incorporate background templates and weighted incremental PCA subspace into the dictionary for sparse representation of target appearance. And algorithm use particle filter as target searching method. In the framework of sparse representation, algorithm use weighted incremental PCA subspace as target part of dictionary and enhance the discrimination of the target model by introducing background templates. These strategies ensure algorithm can better cope with deformation and occlusion situation, and be more resistant to drifting within the process of tracking.In this thesis, testing sequences are released from Visual Object Tracking challenge video collection which is a challenge project in IEEE CS. Experimental results demonstrate that our proposed algorithm outperforms several latest state-of-the-art tracking algorithms including MIL, IVT, STRUCK and so on, in terms of tracking performance.A number of state-of-the-art object tracking algorithms including MIL, IVT, STRUCK and so on, are experimented...