Recovery of Surface Height Using Polarization from Two Views

Abstract

This paper describes a method for establishing stereo correspondences using diffuse polarization information. To do this we exploit the spontaneous polarization of light caused by reflection from dielectric surfaces to recover surface normals. The normals recovered from two different views are used to locally reconstruct surface height. The similarity between reconstructed surface regions determines whether or not a pair of points correspond to each other. The technique is able to overcome the convex/concave ambiguity found in many single view techniques. As the technique relies on smooth surface regions to detect correspondences it is applicable to objects that normally cause difficulty for stereo vision algorithms.