Mean-Shift tracking algorithm based on FCM

Abstract

Concerning the bug of the kernel function bandwidth in traditional Mean-Shift tracking algorithm,a new Mean-Shift tracking algorithm was presented using Fuzzy C-Means (FCM). FCM clustering was used to segment the moving target from background in the YCrCb color-space. According to the rules that the areas of the target in adjoining frames are not changed abruptly,the bandwidth of kernel function was corrected with the statistic of segmented target pixels. Experimental results show that this algorithm can track object accurately and effectively. The sizes of the tracking windows could be adjusted automatically to adapt to the decreasing or increasing sizes of the moving target.