259 Final Project

Physically Based Modeling and Animation of Fire using Graphics Hardware

Proposal

Title

Physically Based Modeling and Animation of Fire using Graphics Hardware

Description

The purpose of this project is to implement the high quality and physically based modeling and animation of fire by Nguyen[Nguyen, 2002] in such a way as to make it more realtime applicable. This will be accomplished by using the features of graphics hardware.

Background and Motivation

Nguyen's[Nguyen, 2002] work produces very nice physically based fire. The target of his work is an animation model which fits inside his photorealistic rendering framework, that is not targeting real-time application. In conjunction with Fedkiw in[Fedkiw, 2001], and [Fedkiw, 2002], they investigate both real-time and non-real-time methods, for both smoke and fire. Using this as background, and additionally building on the grid based graphics hardware work of [Harris, 2002], I intent to produce a real-time implementation of the photorealistic work which is Nguyen's primary focus in [Nguyen, 2002]

---

These images are from the smoke and fire papers by Jensen:

---

Expected Contributions and Accomplishments

• Minimal: a real-time fire renderer, using graphics hardware
• Possibly: all of Nguyen's[Nguyen, 2002] model
• Possibly: some improvement over that model, but this may take beyond the time for this course.

Update

Nguyen's [Nguyen, 2002] model in software. It's now looking to me like this alone is a significant semester project. The model consists of a number of significant elements:

• A temperature/phase function: as the temperature increases, the state chnages from solid fuel to gaseous products of burned fire (not all products may be gas, visible smoke is solid product of the ignition of the fuel. They add this using the smoke work of Fedkiw [Fedkiw, 2001], but it is not the focus of their model, nor of my work with it. So, I'm just using curve such as theirs:
  
• Classic diffusion style 1st order ODE's for temperature and pressure diffusion: This is actually turning out to be a little tricky in the graphics hardware implementation, different equations, variables, and coefficients are used for the different states, so one equation is applied in the fuel regions and another is applied in the burned gas regions.
• Iso-surface evolution: There is an iso-surface that differentiates between solid fuel and ignited gas, corresponding to when the state change occurs. Tracking this isosurface is essential because different equations are applied in the two states. I''m planning on implementing something along the lines of the work that Lefohn and Whitaker did in [Lefohn, 2002] for isosurface evolution in volumes using graphics hardware. I have not attempted this yet, and it appears it alone will be a very significant work.
• Expansion: The local density (both of fuel and of burned gas) must be tracked. The state conversion along the isosurface involves an expansion. The updates in density and pressure and temperature (and also the levelset evolution) depend on the local densities - traditional diffusion style. This expansion behavior is a significant contributor to the "randomness" in the flame paths and motions, and appears to be a factor in convincing fire.
• Color: They use a function of the temperature to determine color. I have not yet figured out how to do this in the graphics hardware.
• Inter-relationships: The temperature affects the density affects the pressure affects the iso-surface affects the state changes affects the expansions effects the temperature affects the ....
• Rendering: They use a photon-mapping with participating medium model. This will not work well for real-time graphics. As it is a volume based model, I'm looking into various hardware volume rendering techniques. This has not been done yet.

In the face of much more to do than I had anticipated, it has been hard to find the best place to start. I have decided to begin by implementing their work in software, then producing something on hardware. This enables me to prototype and experiement with their solution better and faster. It is also very hard currently to debug hardware shader programs. I am nearing completion of the software implementation and have begun drafting plans for the transition to hardware.

I am also considering borrowing directly from the work of Wei [Wei, to appear] and Goodnight [Goodnight, 2003], as both of these use graphics hardware to solve grid based ODEs (Lattice Bolzmann Computations and Boundary Value problems via multigrids, respectively). The multigrid approach particularly is interesting, as it may provide for some significant improvements to the basic fire model. I'm excited to possibly branch into this one, but I have much left to do, possibly too much already.

References

• Fedkiw, Ronald, SIGGRAPH 2002 Course Notes 9: Simulating Nature: Realistic and Interactive Techniques
• Goodnight, Nolan, and Gregory Lewis and David Luebke and Kevein Skadron, A Multigrid Solver for Boundary Value Problems Using Programmable Graphics Hardware, University of Virginia Technical Report CS-2003-03
• Harris, Mark J. and Greg Coombe and Thorsten Scheuermann and Anselmo Lastra, Phsically-Based Visual Simulation on Graphics Hardware, Proceedings 2002 SIGGRAPH/Eurographics Workshop on Graphics Hardware, 2002.
• Lefohn, Aaron and Ross Whitaker, A GPU-Based Three-Dimensional Level Set Solver with Curvature Flow, University of Utah Technical Report UUCS-02-017
• Fedkiw, Ronald and Jos Stam and Henrik Wann Jensen, Visual Simulation of Smoke, SIGGRAPH 2001 Proceedings
• Li, Wei, and Xiaoming Wei and Arie Kaufman, Implementing Lattice Bolzmann Computation on Graphics Hardware to appear in The Visual Computer
• Nguyen, Duc Quang and Ronald Fedkiw and Henrik Wann Jensen, Physically Based Modeling and Animation of Fire, SIGGRAPH 2002 Proceedings.