The mean shift tracking algorithm with adaptive bandwidth based on affine transformation
Lixia Lv, Chenghan Yan · 2015
The kernel bandwidth of the traditional Mean Shift tracking algorithm cannot be changed in real time. It can't achieve accurate tracking when the target size is changing. This paper proposes an algorithm to automatically select the bandwidth of the kernel function, which can achieve accurate tracking when the size of the rigid target is changing. Firstly, it matches the target center of two consecutive frames by using the backward tracking. Then in order to effectively eliminate the false matching and ensure the accuracy of the regression, it extracts and disposes the feature points with regressive calculation on the base of matching. The results of tracking experiment show that the proposed algorithm can achieve accurate tracking in real time when the target size is changing.