An Operational 3D-reconstruction Device using Photometric Stereo and LEDs -- Modeling, Calibration and Resolution

Yvain Quéau (1, 2) , Bastien Durix (1) , Jean-Denis Durou (1) , François Lauze (3)

1 IRIT - Institut de recherche en informatique de Toulouse
2 TUM - Technische Universität Munchen - Technical University Munich - Université Technique de Munich
3 DIKU - Department of Computer Science [Copenhagen]

We describe a photometric stereo-based 3D-reconstruction device. In the classical formulation of the problem, the luminous fluxes are assumed to be directional, which is very difficult to achieve in practice. Rather, we suggest the use of LEDs to illuminate the scene to reconstruct. In doing so, we question the conventional photometric stereo model and must expect its resolution to be more complex. We begin by describing a calibration procedure for the luminous flux emitted by a LED. We then show how to adapt to the new photometric stereo model the classical two-stages resolution method, which involves estimating the albedo and the normals, and then integrating the normals. Then, we reformulate the problem as a system of PDEs, in order to reduce nonlinearities and to eliminate the albedo. We first suggest a fixed point resolution scheme, and then an iterative, ADMM-based scheme. Finally, we extend this study to the case of RGB images.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Yvain Quéau, Bastien Durix, Jean-Denis Durou, François Lauze. An Operational 3D-reconstruction Device using Photometric Stereo and LEDs -- Modeling, Calibration and Resolution. 2016. ⟨hal-01399607v1⟩