Hierarchical Level of Detail Optimisation for Constant Framerate Rendering of Radiosity Scenes

Nirenstein, Shaun and Blake, Edwin and Winberg, Simon and Mason, Ashton (2002) Hierarchical Level of Detail Optimisation for Constant Framerate Rendering of Radiosity Scenes, South African Computer Journal, 29, 23-40.

[img] PDF

Download (1MB)


The predictive hierarchical level of detail optimization algorithm of Mason and Blake is experimentally evaluated in the form of a practical application to hierarchical radiosity. In a novel approach the recursively subdivided patch hierarchy generated by a perceptually refined hierarchical radiosity algorithm is treated as a hierarchical level of detail scene description. In this way we use the Mason-Blake algorithm to successfully maintain constant frame rates during the interactive rendering of the radiosity-generated scene. We establish that the algorithm is capable of maintaining uniform frame rendering times, but that the execution time of the optimization algorithm itself is significant and is strongly dependent on frame-to-frame coherence and the granularity of the level of detail description. To compensate we develop techniques which effectively reduce and limit the algorithm execution time: We restrict the execution times of the algorithm to guard against pathological situations and propose simplification transforms that increase the granularity of the scene description, at minimal cost to visual quality. We demonstrate that using these techniques the algorithm is capable of maintaining interactive frame rates for scenes of arbitrary complexity. Furthermore we provide guidelines for the appropriate use of predictive level of detail optimization algorithms derived from our practical experience.

Item Type: Journal article (paginated)
Uncontrolled Keywords: Rendering, Hierarchical Level of Detail, Hierarchical Radiosity
Subjects: Computing methodologies > Computer graphics
Date Deposited: 26 Sep 2003
Last Modified: 10 Oct 2019 15:36
URI: http://pubs.cs.uct.ac.za/id/eprint/48

Actions (login required)

View Item View Item