Polarized 3D reconstruction for transparent objects based on reflection type separation method

Polarization imaging technique as a simple, fast and low-cost 3D reconstruction method has many applications in the field of 3D reconstruction technology. The principle of polarization 3D reconstruction technique is based on Fresnel's theorem in physical optics, which uses the relationship between the polarization information of the image and the surface normal of the object to reconstruct the object. Since different types of reflections on an object's surface correspond to different polarization information and surface normal, this can lead to significant deviations for transparent objects with a variety of reflection types. Therefore, we used a method that separates the partially polarized diffuse and specular reflection components by applying a polarized industrial camera and a standard industrial camera to acquire images and combining them with a depth estimation method based on polarized images to generate a depth distribution for the surface of a transparent object with a known reflectivity. Non-transparent objects were chosen for technical validation as well as transparent objects for experimental analysis. The obtained results show that compared with the polarized 3D reconstruction method that neglects partially polarized diffuse reflections, this method can reconstruct the surface of transparent objects with sufficient accuracy and has good utility in relevant industrial production.