Técnicas de aceleración para el método de radiosidad jerárquica

  1. Padrón, Emilio J.
Dirigida por:
  1. Margarita Amor Directora
  2. Ramón Doallo Director

Universidad de defensa: Universidade da Coruña

Fecha de defensa: 23 de mayo de 2006

Tribunal:
  1. Emilio López Zapata Presidente/a
  2. Montserrat Bóo Cepeda Secretario/a
  3. Francisco Tirado Fernández Vocal
  4. Xavier Pueyo Sandez Vocal
  5. María Inmaculada García Fernández Vocal
Departamento:
  1. Ingeniería de Computadores

Tipo: Tesis

Teseo: 134339 DIALNET lock_openRUC editor

Resumen

Radiosity is one of the best methods in modelling the physical behaviour of light in a synthetic scene. However, the main drawback is the high requirements in terms of computational and storage costs. Hierarchical radiosity stands out among the different alternatives to reduce complexity in classic radiosity, applying an adaptive subdivision on scene. Hierarchical radiosity still presents, anyway, a high complexity that difficults to process really large scenes. In this work we have developed new solutions for several of the most common bottenecks presented in hierarchical radiosity. Our first goal is to accelerate visibility determination (most consuming task in global illumination), analysing the main existing solutions and proposing a new method based in taking advantage of directional coherence for the rays casted during process. Other aspect we have touched in the thesis is the use of multiresolution models that allow to work with very complex geometrical models in our input scene, isolating geometry detail and illumination detail. Specifically, we have developed a new method to compute hierarchical radiosity based on surface subdivision. Finally, a new parallel solution for computing hierarchical radiosity on multiprocessor systems, allowing huge input scenes is presented. The scene is totally distributed (geometrically and computationally) among the processors in our proposal, and a multi-thread implementation improves the flexibility in the granularity of the parallel execution.