Robust digital image stabilization using the Kalman filter

In this paper, we present a new digital image stabilization (DIS) algorithm based on feature point tracking. Feature points well tracked by the Kanade-Lucas-Tomasi (KLT) tracker are used to estimate the global motion between two consecutive image frames. The motion prediction with the Kalman filter (KF) is incorporated into the KLT tracker to further speed up the tracking process. Moreover, we develop an adaptive KF to better handle the intentional motion of the camera. In addition, a new scheme is also proposed to detect the scene change and automatically change the reference frame. Experimental simulation shows that the proposed DIS algorithm has the characteristics of high accuracy, good robustness to different irregular conditions.