Speaker: René Fürst
Water simulation in computer graphics has been the focus of many research papers in the last years. The proposed algorithms can be classified into offline simulation algorithms and real- time simulation algorithms, which have a big difference in their capabilities. Offline simulation algorithms can mostly simulate water with full detail in 3D because the processing time per frame is not limited very much. Real-time simulation algorithms, on the other hand, impose a hard limit on the processing time per frame and therefore many applications only rely on approximating water with 2D surfaces. While this works well for still water, effects such as breaking waves or simply pouring water into a tank, cannot be simulated. To accomplish this one has to rely on 3D water simulation algorithms, which require a massive amount of processing resources, such that even when using current graphics processing units (GPUs) an optimized algorithm is required.
There are two types of 3D water simulation approaches: particle-based Lagrangian and grid- or mesh-based Eulerian approaches, of which the latter will be the main focus of this thesis. Current real-time capable Eulerian water simulation algorithms rely on many approximations which either cause water to leak without reason or decrease the generality of the algorithm. In this thesis a novel Eulerian fluid simulation algorithm is proposed, which relies on on a octree- based mesh that automatically adapts to any given water geometry. By this it is made sure that thealgorithmisverygeneralandthatitprovidesgoodperformanceinalmostanygivensituation.
The equations evolved in 3D water simulation are discretized with the finite element method in a way, such that the resulting linear equation system can be solved efficiently. Therefore it is possible to let the algorithm run in real-time on current GPUs.