Fixation and a 180 Degree View Simplify Ego Motion Estimation

Although moving human observers actively fixate their eyes on points in the world, computer vision algorithms designed for the estimation of structure-from-motion or egomotion typically do not make use of this constraint. The main contribution of this work is to precisely specify the form of the optical flow field for a fixating observer. In particular, we show theoretically that the use of a hemispherical (retinal) imaging surface generates an optical flow field of a particularly simple form. The predictions of this theory are tested using the first actual hemispherical lens-camera system in computer vision, involving a 180 degree field of view lens. A further contribution is the finding that the sign of flow at the retinal periphery can be used to predict collisions. Comments University of Pennsylvania Department of Computer and Information Science Technical Report No. MSCIS-95-15. This technical report is available at ScholarlyCommons: http://repository.upenn.edu/cis_reports/212 Fixation and a 180 Degree View Simplify Ego Motion Estimation MS-CIS-95-15 GRASP 390 Inigo Thomas Eero Simoncelli Ruzena Bajcsy University of Pennsylvania School of Engineering and Applied Science Computer and Information Science Department Philadelphia, PA 19104-6389