Datenbestand vom 13. März 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 13. März 2019
978-3-8439-3067-3, Reihe Informatik
Anti-Aliasing and Visibility Preprocessing Algorithms for Mobile Graphics Devices
114 Seiten, Dissertation Universität Erlangen-Nürnberg (2017), Softcover, A5
The evolution of computer graphics has reached an astounding level of fidelity, partly because of decades of research but more so because of the advances in hardware development. Modern day graphics cards can host vast arrays of dedicated hardware and provide massive processing powers allowing for very sophisticated light transport simulations. Naturally such hardware is very costly, consumes a lot of power and generates considerable quantity of heat.
An alternative to the high tier hardware are so called mobile GPUs. They provide a cost effective solution and correspondingly limited processing power. The automotive industry has started to implement such computers into the dashboards of their high-segmented products to provide a modern, interactive experience. The creative leeway that follows the adoption of dynamic displays is highly influenced by the fidelity of contemporary generated images that advertise near photorealism but require the capabilities of very costly hardware. The implied challenge is therefore the creation of pleasing images under the constraints of mobile GPUs.
In this dissertation we evolve and combine existing techniques that comply with the restrictions of limited processing powers. We also develop a new approach to pre-compute and store a large portion of the light transport computation. Our first proposal combines a progressive Level-of-Detail technique with a normal map filtering technique to allow a stateless mesh filtering. We address the problems caused by vertex clustering and allow for an aliasing free, high-frequency specular shading.
Following that general-purpose implementation we consider the restriction that are inherent in automotive visualization, namely the determinism of any possible image and animation. We consider the application and compression of Geometry-buffer sequences to allow configurable video playback. Finally we propose a technique to pre-compute and store the visibility information of triangles, for entire camera animation sequences.