Jitter suppression in model-based camera tracking

Model-based camera tracking methods basically require visual cues to be as correctly tracked as possible. However, the requirement is not likely to be satisfied in uncontrolled real environments. Incorrectly tracked visual cues are the main cause of jitter which annoys users in augmented reality. Therefore, we propose a jitter suppression method that guesses the magnitude of camera motion based on a motion blur metric and suppresses jitter in such a way that only the high quality visual cues of which the motion information is not smaller than the magnitude are used for camera pose estimation. For increasing the reliability of the proposed method, existing no-reference edge-based motion blur metrics are evaluated and the most efficient one is selected. Through experimental results with real and synthetic images, we verify the effectiveness of the proposed method.