Abstract
A model-based approach to recognition of glossy objects is presented. Normals of surface patches are obtained by analyzing the polarizational state of the observed rays under illumination of light sources of circular polarization. The object is assumed to lie alone in its stable pose on the floor. Solid models are examined to find the one which matches the observed normals. First, find a candidate model based on relative angles between known surface normals. Then, find the translation so that observed positions of surface normals coincide with those of models in the image. Some examples are also presented.
