An improved adaptive video tracking method based on Mean-shift

摘要

For the mean shift tracking algorithm, we use the histogram feature which contains little information, but the moving direction and velocity information are not been considered, which leads to the target will be missed easily, what is more, the limitation of traditional algorithm which can not change the size of the window to adapt to the size of target, etc. To overcome those weaknesses, we introduce both the target feature representation idea having the features of adapting to window size adjustment and spatial characteristics and the nature of the kernel function, then, there is no need to estimate the probability for all regions. The results of experiment show that compared with using kalman filtering and mean shift algorithm alone, the weighted mean improved filtering algorithm has greatly improved the instability of target tracking and the robustness of the moving target tracking.