Both visualization and simulation tasks make high demands on
accuracy and interactivity. The former is an evident requisite of
any tool focusing on quality, the latter stands for a responsive
system in which the user has real-time control. A permanent
trade-off between both demands can be observed, since high
accuracy is usually time-consuming.
The graphics processing unit (GPU), which evolved to a powerful
general purpose coprocessor during the last decade, is undoubtedly
becoming an ideal instrument for performing visualization and
simulation tasks accurately and in real-time.
One of the main challenges in the context of geometric flow
visualization and fluid dynamics is the representation of motion.
In both fields, particle systems and grid-based structures
constitute basic models describing the objects to which the motion
is applied.
This work is dedicated to the development of GPU-based particle
techniques and their combination with grids in order to improve
the efficiency of fluid simulation and geometric flow
visualization techniques. All resulting algorithms are
characterized by their parallel nature and by being entirely
executable on graphics hardware.
The first aim is to provide an efficient solution to particle
coupling in the context of fluids based on smoothed particle
hydrodynamics. A parallel processing using graphics hardware is
achieved by providing a grid-based mechanism for the efficient
processing of particle neighborhoods.
The second aim is to provide a set of algorithms for various
geometric flow visualization techniques including flow particles,
flow lines and flow surfaces/volumes. The accuracy is explicitly
addressed in all cases. Accordingly, a method for generating
time-adaptive stream and path lines and a special refinement
scheme for streak lines is presented.
The flow volumes, including time surfaces, path and streak
volumes, are based on a parallel reinterpretation of the particle
level set (PLS) method. This reinterpretation is based on a method
for grid-particle interchange which is similar to the one proposed
for particle coupling. Combining a grid and a particle set takes
the advantages from both models: the grid representation is robust
w.r.t. deformations and topological changes, and the particles are
used to reduce numerical diffusion. For the reinitialization of
the level set function, which is the most time-consuming step of
the PLS algorithm, a hierarchical method for computing distance
transforms is proposed. It turns out that the use of a distance
transform is advantageous for realizing a GPU-based PLS framework.