A Novel Tracking Approach Based on Multi-feature Fusion and Feature Space Online Updating

In this paper,we propose a novel tracking approach by combining the local binary pattern(LBP) features with the gray level information under the incremental Fisher linear discriminant to online update feature space and to deal with the problems with illumination changes in object tracking.At the beginning of tracking,to get a more accurate description of the object,expectation-maximization is employed for the segmentation.Then,Monte Carlo method is used to estimate the parameters of feature space based on sampling in the regions of object and background,and the optimal hyperplane is updated.After that,particle Filtering is used to estimate the states of the object.Experimental results show that our proposed approach is effective and outperforms the traditional tracking approaches,such as the color-based particle Filter.