Adaptive model MeanShift tracking
Daihou Wang, Changhong Wang, Zhenshen Qu · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2013
Performance of original color-based MeanShift tracking algorithm decreases drastically under variant illumination environment. To enhance the robustness of the tracking ability under variant illumination environment, an adaptive model MeanShift tracking scheme is proposed in this paper. The statistically approximate LBP texture information is adaptively integrated into the model description to increase the descriptive ability of the model under different illuminating condition. The weighted coefficient of the color information and texture information adjust according to the discriminative ability of the character. Besides, H(Hue) element Gaussian model is introduced for more precisely decription as well as reducing the computational cost of the original histogram-based color model. Experiments on video sequences show the proposed model scheme and advance MeanShift tracking algorithm give effective and robust results in variant illumination condition.