Roland Wahl, Manuel Massing, Patrick Degener, Michael Guthe,
University of Bonn
Institute of Computer Science II.
Römerstraße 164, D-53121 Bonn
Several sophisticated methods are available for efficient rendering of out-of-core terrain data sets. For huge data sets the use of preprocessed tiles has proven to be more efficient than continuous levels of detail, since in the latter case the screen space error has to be verified for individual triangles. There are some prevailing problems of these approaches: i) the partitioning and simplification of the original data set and ii) the accurate rendering of these data sets. Current approaches still trade the approximation error in image space for increased frame rates.
To overcome these problems we propose a data structure and LOD scheme.
These enable the real-time rendering of out-of-core data sets while guaranteeing
geometric and texture accuracy of one pixel between original and rendered
mesh in image space. To accomplish this, we utilize novel scalable techniques
for integrated simplification, compression, and rendering. The combination
of these techniques with impostors and occlusion culling yields a truly output
sensitive algorithm for terrain data sets. We demonstrate the potential of
our approach by presenting results for several terrain data sets with sizes
up to 16k x 16k. The results show the unprecedented fidelity of the visualization,
which is maintained even during real-time exploration of the data sets.