Surface normal estimation of thin transparent objects from polarization of transmitted light

Akihito Watanabe, Daisuke Miyazaki

Abstract: Shape estimation of transparent objects is usually diffi-cult since the light transmits. The proposed method esti-mates the surface normal of objects through polarization analysis of light. Most technique that use polarization for shape estimation analyzes the reflected light. However, in order to observe the reflection over the whole surface, we have to illuminate the object from every direction. In this paper, we propose a novel method for estimating the surface shapes of transparent objects by analyzing the polar-ization state of the transmitted light. The target objects are thin transparent objects, such as bottles or glasses. Thin transparent objects less refract, thus, existing methods that analyze the light transport for shape estimation fail. On the other hand, our polarization-based method can estimate the unique surface normal of thin transparent object.