Log-polar stereo for anthropomorphic robots

Abstract

Stereoscopic vision has a fundamental role both for animals and humans. Nonetheless, in the computer vision literature there is limited reference to biological models related to stereoscopic vision and, in particular, to the functional properties and the organization of binocular information within the visual cortex. In this paper a simple stereo technique, based on a space variant mapping of the image data and a multi-layered cortical stereoscopic representation, mimicking the neural organization of the early stages of the human visual system, is proposed. Radial disparity computed from a stereo pair is used to map the relative depth with respect to the fixation point. A set of experiments demonstrating the applicability of the devised techniques is also presented.