While playing with my planetary terrain renderer, I ran into the problem that that height source generation library that I was using (libnoise) wasn’t giving enough precision – it only provides 8-bit precision for the height values for any given point on the terrain. 256 distinct height values is plenty normally, but it wasn’t really doing the job for me. It was causing nasty ’stepping’ in the terrain, where my LOD would increase, but the additional data that was produced wasn’t fine-grained enough to show any improvement. For example:
Pardon the lack of texturing and so on. Seriously, I’ll get to it after I finish the terrain mesh. You can clearly see the ’steps’ though – the LOD system is doing its best, but without any useful data.
Anyway, I finally got around to rolling my own little perlin noise system tonight. It’s not integrated with the OpenGL renderer yet, but I got some kinda neat effects testing it in a console window. And it produces 32-bit floats 
Warning – if you’re prone to reminiscing about the ‘good old days‘, when real men did everything by command line, you may find this picture invokes a bit of nostelgia. Also, it’s a rather large animated GIF.
EDIT: hmm, so it turns out the GIF is 35mb. Click here to see it, but be warned. Or, there’s a static screenshot below:
I bet everyone thought I forgot about this project. Not so much .
CGL was used for the University Open Day last week to run BSOD on the big display wall, and performed quite well. Photos of that coming as soon as the uni decides to upload some, but in the mean time, enjoy these screenshots.
-
-
BSOD2 running under 5 CGL renders
-
-
A slightly different configuration
It looks a bit more impressive on the big screen
This system is pretty much working now. Here’s it running on live data from the university uplink:
-
-
Typical university traffic
-
-
Toggling some traffic types
-
-
Moving the camera
It still has a small list of changes that need to be made before it goes out to the world, however
As posted previously, I’ve been playing with the beginnings of a LOD terrain renderer for large datasets. I combined this technique with some procedural terrain generation via libnoise, and got some interesting results.
They look pretty awful, but I was more worried about the LOD mechinism than the looks. Rest assured that judicious use of texture splatting and gap fixing would make this look a lot more realistic
I just realised I hadn’t posted any of my shiny new screenshots on here for some time, so prepare for an onslaught of posts
I also took the opportunity to upgrade Wordpress and the k2 theme. Control-F5 may be necessary.
