Hi! First post here, although I've been following this forums for years. I couldn't find anything clear on this topic though, so there it goes.
I'm planning my first "medium scale" 3d game as a hobbyist programmer, an sci-fi space sandbox game, and I'd love to implement "complex" real-scale procedural planets. I've previously done some work with 3d simplex noise and marching cubes, but this seems an step higher. The terrain would be created from a seed and wouldn't be modified so there's no need to save it on disk, generating it on-the-fly instead. But there are other challenges.
My current idea is to use an octree to subdivide the area up to a desired LOD and calculate the density at each point, discarding branches not visible or not belonging to any geometry. In practice, however, an earth-scale planet using a LOD of just 10 meters requires about 21 subdivisions for the most detailed branches. May performance and memory requirements go too high there? Also, mixing geometry from different branches and/or at different LOD levels sounds like a source of gaps and holes in the resulting geometry, putting aside I'm not sure how would LOD perform with tunnels and overhangs. Even worse, if I generate LOD using different frequencies, the branch discarding algorithm would become (even more) complex and error-prone.
As a side note, I'd prefer to do it with C# and XNA 4.0, but if technical requirements arise, I can perfectly develop it on C++. Any suggestions on how could this be done? (Or any related topics I that could have missed?)
Thanks anyway,
Adrián Montesinos
PD: I'm not a native English speaker, so please forgive me for any linguistic errors.
