Object Tracking Algorithm Based On Multi-Feature Fusion And Selection

Object tracking is an important research direction in the field of computer vision. After years of development, illumination variation, occlusion, fast motion, background clutters are still important factors which affect tracking result. Two kinds of object tracking methods based on appearance model are proposed in this paper. They are sparse tracking algorithm based on multi-feature fusion and multi-feature selection tracking algorithm based on support vector machine. Sparse tracking algorithm based on multi-feature fusion makes up the shortcomings of object description with single feature. Every candidate is sparsely represented as a linear combination of all atoms of dictionary. Then the sparse representation model is efficiently solved using a Kernelizable Accelerated Proximal Gradient (KAPG) method. Lastly, in the framework of particle filter, the weights of particles were determined by sparse coefficient reconstruction errors to realize tracking. Multi-feature selection tracking algorithm based on Support Vector Machine which train the classifier with multiple positive and negative samples based on multi-feature descriptions. Make up the shortcomings of the classifier which is trained by the single positive sample. Discriminate all candidate samples with the classifier; choose the candidate sample with the maximum confidence probability as the tracking result. In the tracking step, the incremental subspace learning is introduced to update the templates and positive samples in the two methods. This update strategy considers the change of the appearance model and the original target templates, which can adapt to change in the appearance of the target and avoid the problem of tracking drift. Experimental results show that the two methods can obtain good tracking effect in complex environment.