Combining apparent motion and perspective as visual cues for content-based camera motion indexing

We propose a new method for a qualitative estimation of camera motion from a video sequence. The proposed method suggests to use the properties of vanishing point perspective to complete the information obtained from apparent motion, that is to use a cooperative estimation from several visual cues. Focal length and rotational parameters are first retrieved using perspective, then apparent motion is used to retrieve remaining parameters. The proposed method can retrieve all seven camera motions, including combination of motions. Experimentations confirm that the usefulness of the additional information gained from perspective.