Mean-Shift tracking algorithm with adaptive kernel-bandwidth of target

The bandwidth of traditional Mean-Shift tracking algorithm can not be adapted to size change of target. To overcome this problem, a new adaptive kernel-bandwidth selection method based on the comparison of Bhattacharyya coefficients is proposed in this paper. In this method, the Bhattacharyya coefficient is calculated by using histograms of target image and target background image, and then, a new Bhattacharyya coefficient is calculated according to the candidate image histogram of the current frame and the target background image during the process of tracking. Judging the changing trend of the target by comparing the two former coefficients and the kernel-bandwidth is expanded or shrunk by 10% according to the judgment. Finally, the target area is specified after the calculation above. The simulation results show that the algorithm proposed in this paper verifies the effectiveness of tracking of the targets with changing scales.