PHOTOMETRIC MOTION

Optical flow algorithms typically assume that a surface point's imaged intensity does not change with motion and that geometric distortion of the projected surface shape is the dominant effect of motion. Often, however, the situation is precisely the opposite: Photometric motion, the change in a point's imaged intensity as a consequence of object motion, can easily dominate the other effects of motion. Photometric changes can therefore lead to severe errors in shape recovery if not correctly accounted for. In this paper, it will be shown that photometric motion may be used to obtain a closed-form solution for both surface shape and reflectance, and how this solution may be used to enhance the performance of geometrically based structure-from-motion algorithms is discussed. A simple biological mechanism that accomplishes the recovery of both shape and reflectance is demonstrated.