• Advertisement
Sign in to follow this  

LOD solutions for voxel-generated terrain mesh

This topic is 2140 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts


Despite the topic, I'm not going to ask about stitching holes between LODs or anything like that - it has been discussed many many times already. My problem is that voxel volume, as everyone knows, can have various overhangs, holes and caves. This causes a nasty behaviour when using LOD based on lower resolution, because for more distant LOD a lot of details will be skipped, and where there was cave, there will be rock, hiding the cave until you come very close. This becomes even bigger problem with my water solution, thats also voxel based - it skips whole areas of water where water is very shallow, because of lower sampling rate, which basically means that water should have always highest resolution sampling or it wont look good. Another problem is messed up materials that come from voxel volume information. Different sampling rates produce different materials for surface depending on the sampling step as well.

This missing sampling lead me to a thought - would it be possible to always generate high resolution mesh and somehow make it more coarse for LOD by some kind of processing of mesh vertices? It would solve LOD-holes also, because lower LODs will derive from highest resolution so there won't be differences in vertices - all would depend on algorithm used. This way I could decide in algorithm if given triangles should be or shouldn't be simplified.

So to sum up, my questions are:

a) Is LOD technique not based on sampling but on simplifying highest resolution mesh used and worth looking at?
b) If so, how such technique works, do you have some example algorithms, papers or anything on this that wouldn't be extremely complicated
c) Do you see any disadvantages of this (except computational overhead for generating LODs, but same thing has to happen with sampling method)


Share this post

Link to post
Share on other sites
Hi Noizex,
I did something similar some time ago and I found, that using under-sampling for voxel terrain LOD is actually unusable because
of reasons you mentioned (missing important features etc). I ended up with tessellating voxel representation to highest LOD
and than applied 'standard' mesh reduction technique to generate lower LODs. Here you can see this system in action:

As a good introduction for mesh simplification algorithms you can take a look into Stan Melax's 'Polychop' http://www.melax.com/polychop, looking to quadric error metric LOD stuff is also helpful : http://mgarland.org/...e/cmu/quadrics/ (using QEM you can take material, colors, uvs, etc. into account during mesh simplification process).

This guy has very interesting blog about voxel terrain and other procedural techniques :
http://procworld.blogspot.com/. He explicitly explains lot of techniques you may be interested in also.

P. Edited by born49

Share this post

Link to post
Share on other sites
Sign in to follow this  

  • Advertisement