Size equivalent cluster trees (SEC-Trees) realtime rendering of large industrial scenes

Abstract

In this work we present a rendering method with guaranteed interactive frame-rates in complex 3D scenes. The algorithm is based on an new data structure determined in a preprocessing to avoid frozen displays in large simulative visualizations like industrial plants, typically described as CAD-Models. Within a preprocessing polygons are grouped by size and within these groups core-clusters are calculated based on similarity and locality. The clusters and polygons are building up a hierarchy including weights ascertained within repetitive stages of re-grouping and re-clustering. This additional information allows to choose a subset over all primitives to reduce scene complexity depending on the viewer's position, sight and the determined weights within the hierarchy. To guarantee a specific frame rate the number of rendered primitives is limited by a constant and typically constrained by hardware. This reduction is controlled by the pre-calculated weights, and the viewer's position and is not done arbitrarily. At least the rendered section is a suitable scene approximation that includes the viewer's interests. Combining all this a constant frame-rate including 140 million polygons at 12 fps is obtainable. Practical results indicate that our approach leads to good scene approximations and realtime rendering of very large environments at the same time.