Modeling and rendering of fluid volumes with particle systems

Abstract

Visualization of fluid volumes is necessary in a wide range of applications. Scientific studies of fluid dynamics require the visualization of results of computations. Real?world scenes often contain fluids in the form of water, clouds, smoke, fog, steam, etc. Computer graphics techniques used to generate realistic images of scenes-required for virtual reality and the entertainment industry-must therefore address the problem of modeling and rendering fluids. Fluid modeling is challenging because fluids do not possess well?defined shapes enclosed by surfaces and exhibit dynamic behavior. Physics?based dynamics modeling is often preferred as it involves significantly less effort than “empirical” modeling. Since fluids are often transparent or translucent, rendering them requires capturing light interaction with the fluid volume, rather than only with its surface. The wide range of applications that require fluid visualization, and the variety of approaches developed for fluid modeling, have resulted in several techniques for modeling and rendering fluids. We review these techniques in light of the above issues and develop a framework that enables an easy comparative evaluation of the existing fluid visualization methods. Traditionally, a distinction has been made between texture?based and particle?based techniques in computer?graphics?oriented fluid modeling. The former yield realistic static images but offer limited scope for incorporating physics?based dynamics, while the latter allow easy inclusion of such dynamics but yield images of limited realism. We study these techniques and further group them based on how they visualize the dynamic behavior of fluids. In the framework developed, we identify a new class of techniques that combine texture? and particle?based approaches. We place existing approaches such as the “fluid blobs” technique into this category. Investigating these hybrid techniques is important because they combine advantages of both particles and textures. We next develop a new technique for visualizing fluids, introducing the concept of particle maps. This technique belongs to the hybrid class combining texture? and particle?based methods. Particle maps are kd?tree structures that store the particle system. They allow evaluation of densities at any point within a fluid volume—without using grids—by employing nearest?neighbor information. Use of particle maps helps overcome the limited realism that results from discretizing a fluid volume into particles, and eliminates artifacts produced when grids are used to convert particle systems into densities. An important property of a fluid is its dynamic behavior. We adapt to computer graphics the fully Lagrangian formulation of the Vortex Element Method (V.E.M.) for simulating fluid dynamics. Both two?dimensional and three?dimensional V.E.M. are studied, and methods suitable for computer graphics are adopted. The particle?maps data structure is then exploited to propose an approximation to the V.E.M. suitable for animations. We design the rendering algorithm so that the nearest?neighbor information obtained during rendering is reused for dynamics computation. Thus, we eliminate the explicit dynamics?computation phase present in all existing animation techniques for fluids. This creates a tie?up between dynamics computation and rendering resolution, which can be exploited for level?of?detail modeling. The V.E.M.?based simulation also reduces the effort required to model wind fields that control fluid motion. We develop techniques for rendering both low?albedo and high?albedo fluid volumes modeled by particle maps. For rendering low?albedo fluids, we make efficient use of range searches on particle maps to evaluate ray attenuation. Rendering high?albedo fluids is achieved by merging the photon?map technique with particle maps. We extend the concept of particle maps into the time dimension to incorporate temporal information into individual animation frames. A technique of spacetime ray tracing of four?dimensional particle maps is proposed for generating animations of fluid volumes. This technique significantly reduces the number of particles required to model a gaseous volume. We demonstrate the effectiveness of the proposed technique with images and animations. We expect the framework, tools, and techniques developed in this thesis to have a significant impact on fluid modeling and rendering for scientific visualization, virtual reality, and entertainment applications.