Catadioptric Self-Calibration

Abstract

We have assembled a standalone, movable system that can capture long sequences of omnidirectional images (up to 1,500 images at 6.7 Hz and a resolution of 1140/spl times/090). The goal of this system is to reconstruct complex large environments, such as an entire floor of a building, from the captured images only. In this paper, we address the important issue of how to calibrate such a system. Our method uses images of the environment to calibrate the camera, without the use of any special calibration pattern, knowledge of camera motion, or knowledge of scene geometry. It uses the consistency of pairwise tracked point features across a sequence based on the characteristics of catadioptric imaging. We also show how the projection equation for this catadioptric camera can be formulated to be equivalent to that of a typical rectilinear perspective camera with just a simple transformation.