NASA’s Mars Rover mission planners find themselves with staggering volumes of terrain data from orbital and surface-based assets at vastly different scales, making navigation planning problematic and error-prone. Current state-of-the-art tools for terrain visualization, such as Google Earth, only manage terrain represented as Digital Elevation Maps (not full 3-D), and integrate high-resolution imagery, but not high-resolution or 3-D terrain geometry. In addition, multiple sources of information are presented as image layers, not as a unified surface.
DigitalFish, together with researchers from Carnegie Mellon University, secured an STTR grant to advance the tools for planetary terrain mapping to address these two prime areas. DigitalFish proposed using subdivision surfaces (subdivs), a compact, multiresolution representation for 3-D surface geometry in computer graphics. An advantage of subdivs over other 3-D formats such as NURBS is that subdivs support arbitrary topology: the subdiv base mesh may freely intermix faces and vertices of different valence. This topological flexibility is an advantage in the presence of large local variations in mesh density.
We developed two novel techniques to address these problems: terrain reconstruction from volumes, and a unified mesh representation.
A central contribution of this work is a procedure for converting terrain data from disparate sources into a volumetric representation from which a unified mesh is reconstructed. This method provides fine-grain control over the accuracy of the reconstructed surface and supports any number of sources with overlapping terrain features. We use DreamWorks’ OpenVDB for volume processing currently. We plan to extend the processing to run on scaled GPU clusters such as NVIDIA Grid.
We augment Pixar’s subdivision surfaces by allowing nested control meshes within the subdivision hierarchy. The resulting hierarchical subdiv mesh is a graph-theoretic tree that defines a subdivision surface at each node. Each child node is a finer resolution than its parent and occupies a “hole” which it smoothly interpolates along their shared boundary to guarantee a crack-free surface. Because each hole is simply represented by metadata in the parent’s subdivision hierarchy, no mesh surgery occurs that could disturb the terrain data and introduce error.
Subdivision surfaces are supported in a recently released open-source library, Pixar’s OpenSubdiv (OSD). A key advancement within OSD is the reformulation of the core subdivision algorithms for fast evaluation on large- scale Single Instruction, Multiple Data (SIMD) architectures, including on Graphics Processing Units (GPUs) found in desktop and mobile computing devices
We hope next to build a prototype software tool enabling NASA mission planners to incorporate these terrain-mapping improvements into their ongoing mission planning. This work was selected by NVIDIA's GPU Technology Conference as a GTC poster. The GTC Poster shows further details. Beyond mapping Martian terrain, we see other uses: