High-Perfomance Algorithms for Real-Time GPGPU Volumetric Cloud Rendering from an Enhanced Physical-Math Abstraction Approach

  1. JIMÉNEZ DE PARGA BERNAL QUIRÓS, CARLOS
Dirigida por:
  1. José Antonio Cerrada Somolinos Director
  2. Sebastián Rubén Gómez Palomo Codirector

Universidad de defensa: UNED. Universidad Nacional de Educación a Distancia

Fecha de defensa: 23 de octubre de 2019

Tribunal:
  1. Raquel Dormido Canto Presidenta
  2. Francisco Javier Cabrerizo Secretario/a
  3. Gonzalo Pajares Vocal

Tipo: Tesis

Resumen

The current trend in computer graphics applications requiring landscape rendering, such as flight simulators, computer games and educational software, is implementing realistic outdoor scenarios with a smooth interactive experience. Achieving real-time performance requires powerful hardware to render the details of the landscapes in execution time. One of the features that contributes most to realism is a cloudy sky with credible lighting and the possibility of approaching, moving around and traversing the clouds. Since the upcoming of computer graphics in the early eighties, engineers, physicists and mathematicians like James Blinn, James Kajiya and Geoffrey Gardner have tried to recreate clouds analytically with both low and high rendering detail. The low-detail models were ren- dered in real-time in low-resolution work-stations with straightforward geometry; whereas the highly detailed ones required considerable rendering hours in those ages due to the physically oriented models with all meteorological equations. Nowadays, engineers demand the second type in real-time. Achieving this feat with those models requires massive multi-core graphics hardware that is used in the animation industry but it is expensive. The goal of the present research is achieving a similar cloud quality with high performance using consumer-level hardware. The research follows the ontogenetic approach that performs cloud rendering using a high-level description to avoid the heavy calculations of physical- mathematical models. In the implementation of the ontogenetic method, the first problem that this thesis resolves is the static rendering of the gaseous mass and the cloud shapes. The key contribution of this part is using modern techniques of noise sampling along with Gaussian-optimized primitives or randomized fractal clouds described in a formal language. The use of these novel techniques to reproduce the irregular nature of cumuliform clouds and the effect of light-scattering with low computing cost was crucial to achieve a quality level similar to other high-computing methods. The second problem that this thesis deals with is the dynamic behaviour of cloud masses when they are affected by pressure and wind advection. A novel utilization of multi-core hardware for the simulation of the atmospheric fluid together with simplified algorithms of animation improve the realism in cloud deformation and translation. In summary, this thesis demonstrates, from empiric and image quality benchmarks, that it is possible to accomplish real-time cloud rendering without expensive hardware elements and with an optimum balance between realism and performance. Besides, the actual implementation provides a reusable framework for the graphics industry.