CVSL Logo
FrancaisHome
AboutPeopleResearchPublicationsEventsProfile
About

 

 

 

CERVIM

REPARTI

MIVIM

Realistic Visual Simulation of an Environment: Projective Reconstruction and Extraction of Texture Properties from a Limited Number of Photographs.
Masters
Marc Bédard
Patrick Hébert (Supervisor)
Problem: Photogrammetry software packages enable the reconstruction of 3D objects or environments from an ensemble of photographs. These software packages also allow the application of textures extracted from the photographs. Since the ensemble of photographs can include several viewpoints of the same texture, it is the texture in the image whose viewpoint corresponds the best with the point of observation of the reconstructed model which is applied during the rendering [Debevec]. However, varying the illumination of the textured model according to the approach does not give a realistic rendering. To increase the realism, the textures must be characterized and their properties must be extracted.
Motivation: A method has been developed to extract the radiance map of an environment from photographs [Debevec and Malik]. This method was used to characterize the textures of the model by means of pseudo-BRDFs [Yu and Malik]. It is thus now possible to simulate lighting changes and to obtain a realistic rendering. However, although interesting, the approach of Yu and Malik requires several human interventions at various steps of the extraction of texture characteristics, such as the delimitation of the shadow zones in photographs. The goal of this project is to facilitate the use of modeling techniques and to make them accessible to all.
Approach: To improve the approach of Yu and Malik by automating manual interventions. The first automation to be achieved consists in choosing the regions of low intensity variance. This selection would enable the radiance to be considered constant, which would minimize the blurriness induced by the optical system, and also reduce the complexity of the computations. Another intervention which should be automated is that of the selection of the areas characteristic of the sky so as to proceed to the fitting of the revised sky model. Finally, we will attempt to limit human intervention during the extraction of shadow areas.
Challenges: The real challenge of automation is to ensure the robustness of the process.
Applications: Applications are numerous. A tool of this type is in great demand in the areas of cinema, virtual or augmented reality, video games, architecture, etc. since this would enable the rapid, easy and realistic reproduction of large 3D environments.
Calendar: January 2003 to April 2004
Last modification: 2004/01/30 by mbedard

     
   
   

©2002-. Computer Vision and Systems Laboratory. All rights reserved