Simulation d'éclairage dans des environnements architecturaux complexes : approches séquentielle et parallèle

Daniel Meneveaux 1
1 SIAMES - Computer generated images, animation, modeling and simulation
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, INRIA Rennes
Abstract : Computing global illumination for complex environments in moderate time and walking through them is one of the challenges in computer graphics. Indeed, hierarchical radiosity is a very demanding process in terms of computation time and memory ressources even for scenes of moderate complexity. A preprocessing is then necessary. This preprocessing consists in partitioning the environment into regions (called cells) and determining visibility between these regions. We propose a new partitioning method that relies on image analysis and consists in matching cell models with geometric elements within the scene. Compared to the binary space subdivision technique, our method results in a low number of cells fitting at best with the environment topology. Radiosity computation is performed only for a resident subset of cells which changes during the resolution process. The geometric and photometric information is stored on the disk. Lighting simulation is then considered as the composition of elementary tasks consisting in (i) loading in memory the necessary information for the simulation, (ii) performing radiosity computations, (iii) updating the database located on the disk. However, in order to reduce the numerous disk transfers, the computations must be ordered to make the radiosity algorithms tractable. To this end, we propose several strategies making use of the cells resulting from the scene partitioning as well as a graph expressing visibility between these cells. These strategies predict for the short, medium and long range terms the disk transfer costs that determine the choice of the cells that will shoot their energy. This algorithm is the starting point for our SPMD parallel hierarchical radiosity program, based on a public domain software called MPI (Message Passing Interface). In this case, the database is stored on a single disk and accessed by all the processors (through NFS in case of a network of computers). Each processor performs computations for a subset of regions according to the same ordering strategies as our sequential algorithm. Dynamic load balancing relies on a task stealing approach while termination is detected by configuring the processors into a ring and moving a token round this ring.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00007237
Contributor : Daniel Meneveaux <>
Submitted on : Thursday, October 28, 2004 - 9:47:22 AM
Last modification on : Friday, November 16, 2018 - 1:25:06 AM
Long-term archiving on : Friday, April 2, 2010 - 9:02:10 PM

Identifiers

  • HAL Id : tel-00007237, version 1

Citation

Daniel Meneveaux. Simulation d'éclairage dans des environnements architecturaux complexes : approches séquentielle et parallèle. Interface homme-machine [cs.HC]. Université Rennes 1, 1998. Français. ⟨tel-00007237⟩

Share

Metrics

Record views

308

Files downloads

598