Full text
| |
| Author(s): |
Rafael Felipe Veiga Saracchini
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2012-04-18 |
| Examining board members: |
Jorge Stolfi;
Jose Ricardo de Almeida Torreão;
Roberto Marcondes Cesar Junior;
Hélio Pedrini;
Anamaria Gomide
|
| Advisor: | Jorge Stolfi |
| Abstract | |
The main goal of this PhD thesis is the development and implementation of innovative techniques for stereo vision or stereo - the reconstruction of a tridimensional model of a scene from two or more two dimensional images. Specifically, we studied and improved photometric stereo methods, which use images taken from the same viewpoint under different lighting conditions, and determine the surface normal of the scene at each pixel. We mainly studied methods that determine the lighting conditions and surface finish by the analysis of images of reference objects (gauges or example objects) with known geometry. The contributions of this thesis consist of: (1) development of virtual example objects in order to eliminate noise and defects present in images of real example objects; (2) a 2D hashing grid method for fast normal computation; (3) a robust approach based in Bayesian inference for surface normal computation in the presence of shading anomalies, and (4) two methods for multi-scale normal map integration that can be used in maps with missing or noisy data. We also investigated a practical application of such methods: the capture of the 3D facial geometry of a person in movement (AU) | |