Target tracking is a hot research field of computer vision which is widely used in many fields, such as intelligent navigation, video security, arms manufacturing etc. In recent years, along with the successful application of the sparse representation in the field of signal processing, many scholars also apply the sparse representation to target tracking. This object tracking method of combined sparse representation maintains good robustness when dealing with light and occlusion. However, the traditional target tracking algorithm based on sparse representation doesn’t work under all conditions. Therefore, in this paper, in-depth study of the sparse representation has been made in order to improve the real-time and robustness of target tracking algorithm in complex scenes. The main contents of this paper are as follows:1. Studied in compressive sensing theory including sparse representation, sparse decomposition and the design of dictionary. For unknown sparsity of sparse decomposition algorithm, proposed an adaptive sparsity reconstruction algorithm. This method combines the idea of adaptive sparse and segmentation orthogonal matching tracking algorithm so as to ensure that the sub-orthogonal matching tracking algorithm can accurately reconstruct images under unknown sparsity.2. Studied gabor feature the design of dictionary based on sparse representation. Proposed a target-tracking algorithm based on gabor dictionary sparse representation in order to deal with illumination changes and attitude changes of objectives by using gabor function to extract target template features. The method establishes gabor dictionary by using the original frame features of target template and then makes sparse representations of candidate targets while tracking by gabor dictionary.3. Studied a particle filter tracking algorithm based on sparse representation by using particle filter tracking algorithm. Proposed a target-tracking algorithm based on sparse learning so as to deal with occlusion problems when tracking. This method uses the block sparse learning based on traditional L1 trackers. By making full use of the priori information of occlusion for learning, quick reconstructions of the target is ensured when the occlusion is not sparse.In order to verify the effectiveness of the proposed algorithm, the algorithm is tested using download data through network in which uses sparse representation describe tracking-target and makes target templates and occlusion templates based on Gabor feature templates. Tracking targets by particle filter, decisions of occlusions are made through sparse representation coefficient; occlusions are used to sparse learning at the same time. Experiments show that the proposed algorithm can effectively track moving targets of large occlusion area and has robustness in illumination changes and pose changes. |