Central Panoramic Cameras, Design, Geometry, Egomotion

A Dissertation Presented to the Faculty of Electrical Engineering of Czech Technical University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

The whole PostScript file compressed by gzip.

Abstract: The thesis desribes contribution to basic research on panoramic vision. The solid methodology for a design of a hyperbolic mirror has been derived. Mirror parameters are mainly restricted from several requirements: camera used, sufficient field of view, practical realization and problems of focusing. Our methodology has been used in practice. The epipolar geometry for a pair of central panoramic cameras is the main novel contribution of this thesis. This geometry constraints positions of corresponding points. An epipolar conic is assigned in the second image to each point in the first panoramic image and vice versa. The epipolar geometry serves as a basis for the motion computation and scene reconstruction algorithms. Our results give a general framework for a pair of the panoramic cameras that preserve single effective viewpoint. Our method for a robust estimation of the epipolar geometry is based on a known statistical theory. The main novelty is the using diagonal elements of the essential matrix as the quality parameter of an estimated epipolar geometry. Our algorithm can cope with outliers and it is faster than iterative algorithms. Experiments with egomotion computation has verified superiority of panoramic cameras in this field. The derivation of the optimal point normalization is our ongoing research.
Keywords:
omnidirectional vision, epipolar geometry, panoramic cameras, hyperbolic mirror, stereo, catadioptric sensors.