Stereo Matching Validation using Spherical Image Mapping

As the increasing number of cameras adopt fisheye lenses for sensing large area at once, an effective stereo vision algorithm is required for fisheye images. The exterior part of the fisheye image is distorted so that the existing planar algorithm may not work properly. In this paper, we propose a method in which the fisheye image is mapped to a sphere and the feature detection is performed on the sphere. With experimental results, the proposed algorithm for the spherical image shows better performance than the existing planar algorithms.