Tez No	İndirme	Tez Künye	Durumu
400397		Efficient surface reconstruction for SPH fluids / Yazar:GİZEM AKINCI Danışman: PROF. DR. MATTHIAS TESCHNER ; PROF. DR. JAN BENDER Yer Bilgisi: Albert-Ludwigs-Universität Freiburg im Breisgau / Yurtdışı Enstitü / Bilgisayar Grafiği Ana Bilim Dalı Konu:Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol = Computer Engineering and Computer Science and Control Dizin:	Onaylandı Doktora İngilizce 2014 104 s.
Smoothed Particle Hydrodynamics (SPH) is a popular particle-based fluid simulation technique with its wide application area ranging from the entertainment industry (e.g., movies, video games, commercials) to scientific research fields (e.g., medical simulations, dam and flood simulations, petroleum research). As the method has been gaining increased interest, simulation scenarios have been becoming more diverse and more complex. However, surface extraction from particle-based fluid data still remains as one of the bottlenecks due to large computation time and memory consumption requirements, and also because of difficulties in achieving smooth and detailed surfaces. This thesis presents some new techniques that focus on achieving high quality SPH surfaces efficiently. The thesis starts by reviewing techniques regarding the simulation and surface reconstruction for SPH fluids. Those surface reconstruction techniques predominantly focus on essentials of the field, e.g., lowering memory consumption and computation time, increasing the surface quality or both. Next, we present a novel parallel algorithm so as to improve the performance of time consuming scalar field construction step. Our method scales nearly linearly on multi-core CPUs and up to fifty times faster on GPUs compared to the original scalar field construction approaches. The method scales with the fluid surface instead of its volume by constructing the scalar field only in the narrow band region where the fluid surface actually lies. Therefore, the method works efficiently even on single-core CPUs. Furthermore, we present some values for an efficient parameter setup which affect the final quality of the reconstructed surface. In our next contribution, we focus on achieving smooth and detailed fluid surfaces within a reasonable time. Thus, we propose to use a method, based on post-processing of surface meshes, that is applicable to particle position data sets in a frame by frame basis. Our method combines existing scalar field computation techniques with two post-processing steps: decimation and subdivision. This is motivated by an improved representation of smaller surface details, reduction of bumps in flat regions, reduced overall computation time and reduced memory consumption. Our results demonstrate that in comparison to other approaches with comparable surface quality, our pipeline runs up to twenty times faster with up to 80% less memory and secondary storage consumption. Finally, we present an adaptive spatial data structure, which adapts its cells according to the surface curvature. Low curvature regions are handled in low resolution cells with fewer triangles, while high curvature regions are represented with more triangles using high resolution cells, which helps to preserve surface details. Mesh blocks from different resolution cells are seamlessly stitched by closing cracks with new triangles. Our method produces similar results with less number of triangles, up to four times better memory consumption, and up to 60% better performance when compared to the single level high resolution uniform grid approach. Throughout this thesis, we present various comparisons of our methods with previous approaches in the literature in order to highlight the benefits of our approaches. Visual comparisons and performance comparisons are given at the end of each corresponding chapter together with their explanations.