Abstract: An image stabilization algorithm based on characteristic point tracking is presented, which abstracts a group of characteristic points from the reference images, tracks the characteristic points based on Kalman filtering in the follow-up images, and then matches coordinates of the characteristic points in current images and reference images. The parameters of displacement and rotation are derived by the affine model, the matrix of images is converted reversely and the stable video images are obtained. The result from experiment indicates that the algorithm which is easy to operate and has great ability to resist any disturbance can get good effect of stabilization.