A binocular stereo algorithm for log-polar foveated systems

Abstract

Foveation and stereopsis are important features on active vision systems. The former provides a wide field of view and high foveal resolution with low amounts of data, while the latter contributes to the acquisition of close range depth cues. The log-polar sampling has been proposed as an approximation to the foveated representation of the primate visual system. Although the huge amount of stereo algorithms proposed in the literature for conventional imaging geometries, very few are shown to work with foveated images sampled according to the log-polar transformation. In this paper we present a method to extract dense disparity maps in real-time from a pair of log-mapped images, with direct application to active vision systems.