Generalized Scene Reconstruction

The feasibility of new passive approach to 3D reconstruction, called Generalized Scene Reconstruction (GSR), is explored in this paper. Generalized scenes are defined to be "boundless" spaces that include non-Lambertian, partially transmissive, textureless and finely-structured matter. GSR enables such scenes to be effectively reconstructed by devicesusing-scene-reconstruction (DSRs) such as mobile phones, augmented reality (AR) glasses and drones. A new data structure called a plenoptic octree is introduced to enable efficient (database-like) light and matter field reconstruction in DSRs. To satisfy threshold requirements for reconstruction accuracy, scenes are represented as systems of partially polarized light interacting with matter. To demonstrate GSR, a prototype imaging polarimeter is used to reconstruct, by sensing and modeling generalized polarimetric light fields, highly reflective, haildamaged automobile body panels. Follow-on GSR experiments are described. Keywords—Inverse Light Transport, Volumetric Scene Reconstruction, Non-Lambertian Scene Reconstruction, Visual SLAM, Imaging Polarimetry.