Shape from polarization in the far IR applied to 3D digitization of transparent objects

Abstract

This paper presents an application of "shape from polarization" method in the far Infrared range with applications for three-dimensional reconstruction of transparent objects. Shape from polarization is a recent application of more general polarization imaging technique having the aim to digitize the shape of the observed object. The principle is to evaluate the normal on each observed point followed by an integration procedure. The technique is well developed in the visible domain, but not in the far infrared domain due to the requirement of telecentric optics. We propose here a complete setup in the 8-13 micrometer spectral band with an appropriate source and a reconstruction method including the pinhole camera model in order to use standard optics for the camera. We present primary results of threedimensional digitization of transparent objects.Keywords: Polarization, thermal imaging, transparent objects, long wave infrared and three-dimensional digitization State of the artRecovering three-dimensional surface shape of objects, remains an important research area of computer vision. However, when dealing with the inspection of transparent or highly reflective surfaces like mirror, standard solutions are not available yet. Referring to these non "Lambertian" surfaces, Ihrke et al [1] published an exhaustive survey which was recently completed by Meriaudeau et al.[2]for transparent objects.Among the recent works, approaches such as "scanning from heating" [3], where the three-dimensional data is reconstructed from a heat pattern visualized with a calibrated IR camera or "shape from induced fluorescence" [4] where the three-dimensional reconstruction takes place thanks to a generated visible pattern induced by florescence onto the object surface, appear to be very promising for three-dimensional inspection with potential adaptation for transparent as well as specular objects.The shape from polarization technique, based on the analysis of polarimetric property of the light reflected from the object, is also well adapted for non "lambertian" surfaces. This is an unconventional approach because it does not rely on the tradionnaly used parameter in machine vision such as the intensity and wavelength. A standard camera model limited by brightness and hue is not enough to measure the polarization parameters.In this paper, we used the shape from polarization method in the far infrared band. This spectral band corresponds to a part of the spectrum where transparent materials like glass and some plastics appear opaque. Since the pioneer works of Wolff [5] on polarization imaging, recent extensions have been made possible by active lightings systems [6], multispectral approach [7] and infrared imaging. In the field of remote sensing Middle IR band is then used for material detection and separation [8]. The last example used the near IR band [9] in order to build a depth map of the outdoor environment. This brings new interest in polarization imaging techniques. Proposed MethodThe acquisition and reconstruction proc...