Controlling memory consumption of hierarchical radiosity with clustering

Memory consumption is a major limitation of current hierarchical radiosity algorithms, including those using clustering. To overcome this drawback we present a new algorithm which reduces the storage required for both the hierarchy of subdivided elements and the links representing light transfers. Our algorithm is based on a link hierarchy, combined with a progressive shooting algorithm. Links are thus stored only when they might transfer energy at subsequent iterations. The push-pull and refine/gather steps of hierarchical radiosity are then combined, allowing the simplification of subtrees of the element hierarchy during refinement. Subdivided polygons replaced by textures and groups of input objects contained in clusters may be deleted. A memory control strategy is then used, forcing links to be established higher in the link hierarchy, limiting the overall memory used. Results of our implementation show significant reduction in memory required for a simulation, without much loss of accuracy or visual quality.