FAST Corner Detector in Discrete Spherical Image

Abstract

It is general to use a fisheye camera for obtaining a full view image, but the images obtained by a fisheye camera are not rotation invariant, that is, the distortion of scenes on the image varies with their place of image. On the other hand, the content of a spherical image is invariant with the camera rotation and scenes do not vary. In this paper, we propose to detect corners on spherical images. We map a full view image to a discrete spherical image for obtaining a rotation invariant spherical image, and modify the conventional FAST (Features from Accelerated Segment Test) corner detector for directly applying it to the discrete spherical image.