Open AccessPassive non-line-of-sight imaging using plenoptic information
Author(s) -
Di Lin,
Connor Hashemi,
James R. Leger
Publication year - 2020
Publication title -
journal of the optical society of america. a, optics, image science, and vision./journal of the optical society of america. a, online
Language(s) - Uncategorized
Resource type - Journals
SCImago Journal Rank - 0.803
H-Index - 158
eISSN - 1520-8532
pISSN - 1084-7529
DOI - 10.1364/josaa.377821
Subject(s) - optics , artificial intelligence , computer vision , perspective (graphical) , computer science , scattering , homogeneous , surface (topology) , spatial frequency , physics , line of sight , mathematics , geometry , thermodynamics , astrophysics
We present a methodology for recovering the perspective imagery of a non-line-of-sight scene based on plenoptic observations of indirect photons scattered from a homogeneous surface. Our framework segregates the visual contents observed along the scattering surface into angular and spatial components. Given the reflectance characteristics of the scatterer, we show that the former can be deduced from scattering measurements employing diversity in angle at individual surface points, whereas the latter can be deduced from captured images of the scatterer based on prior knowledge of occlusions within the scene. We then combine the visual contents from both components into a plenoptic modality capable of imaging at higher resolutions than what is allowed by the angular information content and discriminating against extraneous signals in complex scenes that spatial information struggles to discern. We demonstrate the efficacy of this approach by reconstructing the imagery of test scenes from both synthetic and measured data.