Title :
Mixing catadioptric and perspective cameras
Author_Institution :
INRIA Rhone-Alpes, Montbonnot, France
Abstract :
We analyze relations that exist between multiple views of a static scene, where the views can be taken by any mixture of para-catadioptric, perspective or affine cameras. Concretely, we introduce the notion of fundamental matrix, trifocal and quadrifocal tensors for the different possible combinations of these camera types. We also introduce the notion of plane homography for mixed image pairs. Generally speaking, these novel multi-view relations may form the basis for the typical geometric computations like motion estimation, 3D reconstruction or (self-) calibration. A few novel algorithms illustrating some of these aspects, are described, especially concerning what we call calibration transfer, using fundamental matrices, and self-calibration from plane homographies.
Keywords :
cameras; computer vision; image reconstruction; motion estimation; 3D reconstruction; catadioptric cameras; fundamental matrix; mixed image pairs; motion estimation; perspective cameras; plane homography; quadrifocal tensors; self-calibration; trifocal tensors; Calibration; Cameras; Equations; Geometry; Image reconstruction; Layout; Mirrors; Motion estimation; Robustness; Tensile stress;
Conference_Titel :
Omnidirectional Vision, 2002. Proceedings. Third Workshop on
Print_ISBN :
0-7695-1629-7
DOI :
10.1109/OMNVIS.2002.1044489