A Robust Tracking Method Based on Two-Dimensional Histogram and Generalized Divergence

Due to its robustness and computational efficiency, the mean shift based algorithms have achieved considerable success in object tracking. Color histogram is used in these algorithms to represent the target model. The Bhattacharyya coefficient is employed as the similarity measure to evaluate the difference between the model and the target candidates. In practice, there exist some limits in these approaches. First, images with very different appearances can have similar histograms for the lack of spatial information. Second, the Bhattacharyya coefficient is not quite discriminative, especially in higher dimensions. In this paper, the color information and other spatial features (such as texture etc) are integrated to construct a two-dimensional histogram to represent the target model. To achieve more accuracy of tracking and robustness to background motions, generalized divergence is introduced as a similarity measure. The robustness and accuracy of the present method is demonstrated in the experiments.