Dynamic Updating of Planar Structure and Motion: The Case of Constant Motion

摘要

This paper describes an algorithm which uses the formalism of the extended Kalman Filter to provide a dynamic solution to the nonlinear reconstruction of the 3D structure and motion of a planar facet moving with arbitrary but constant motion relative to a single camera. By integrating measurements of visual motion over time, the algorithm restores the coupling between the scene structure and rotational motion which is absent in instantaneous or “snapshot” processing of visual motion, thereby disambiguating the two possible values of the rotational motion which arise in such processing. The dynamic recovery of planar structure from motion is demonstrated on real imagery in sequences of several tens of frames taken both indoors in a well-controlled environment and outdoors from a moving vehicle.