Toward topological localization with spherical Fourier transform and uncalibrated camera

Abstract

The paper presents a new metric for computing the image similarity using the Spherical Fourier Transform (SFT) of omnidirec- tional images projected on a sphere. This metric is designed for image- based localization of mobile robot. Given a set of omnidirectional images, generated from an uncalibrated catadioptric sensor composed of an hy- perbolic mirror and a prospective camera, we map them on the sphere to calculate the Spherical Fourier Transform. The transform coefficients are used to calculate the similarity between two images, and the rotation around optical axis of the camera is calculated using the SO(3) transform (SOFT).