Processing differential image motion

Abstract

The inference of three-dimensional camera motion parameters and the layout of a scene from image flows becomes particularly simple from a computational point of view if the scene contains depth variations. Under this condition, the differential image motion yields a simple estimate of the translation field lines at image locations corresponding to depth discontinuities in the scene. This in turn facilitates closed-form solutions of camera motion parameters and environmental depth. Our results may have relevance to human motion perception, which also seems to rely on depth variation in processing image motion.