Improvement of tracking capability in the mean shift tracker

In order to resolve the poor performance of color-based tracker for the fast rotating object in low-resolution vision, the search region of the tracker needs enlarging to a great extent, and the tracking loop requires setting to several times in very frame. Moreover, the states of the tracked object in high-resolution vision can be estimated from its incomplete observations in low-resolution vision by using the Kalman filter based on the optimized state-space model (SSM) identified by the maximum likelihood (ML) estimator.