Acquisition and analysis of dense 3D geometry and reflectance
File(s)
Author(s)
Kampouris, Christos
Type
Thesis
Abstract
Acquiring the 3D geometry of a surface has been a very active area of research in computer vision. While passive acquisition methods, like multiview stereo, have been the most common, they obtain the coarse 3D geometry and don’t recover fine-scale details. Photometric stereo, on the other hand, has the ability to obtain high detailed results but requires a complex capturing setup with a large number of light sources. In this thesis we explore practical photometric stereo systems to obtain high resolution geometry and reflectance. First, we present a mini photometric stereo sensor using affordable, off-the-shelf components and designed to be used in robotics. The sensor provides a fast and easy way to acquire dense normal maps and albedos in the field. We used the sensor to collect a large dataset of fabrics and propose a novel approach to material classification using handcrafted and learned features. Next, we present a multispectral light stage consisting of 168 pairs of RGB and white lamps. We describe the design and implementation of the light stage and explain the decisions we made to facilitate the construction. We show that it provides a versatile solution for a variety of applications including reflectance capture, image-based lighting reproduction, and multiview facial geometry and appearance acquisition. Further, we propose a novel method for view-independent diffuse-specular separation of albedo and photometric normals using binary spherical gradient illumination, without requiring polarization of the light stage. We also demonstrate an efficient two-shot capture using spectral multiplexing of the illumination that enables diffuse-specular separation of albedo and heuristic separation of photometric normals.
Version
Open Access
Date Issued
2018-06-27
Date Awarded
01/05/2024
License URL
Advisor
Zafeiriou, Stefanos
Ghosh, Abhijeet
Sponsor
European Commission
Engineering and Physical Sciences Research Council
Grant Number
FP7-ICT-288553 CloPeMa
EP/N006259/1
Publisher Department
Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)