So what's the progress since the last blog entry?
All generated vegetation is nearly done, after we totally reworked the scripts for conifer trees for better results. There will just be a final color tuning on the vegetation. You can see some ingame screenshots showing how it looks now. There are some small issues with multiplication of sections when trees are blended, problem is in some trunk lightmaps that are not shared through LODs, something easy to fix. But the main focus is now on special vegetation LODs. We also found problems with LOD switching on bushes, because LODs were too different, we reworked the algorithm for choosing representative polyplanes and made this algorithm more controllable and flexible, there was also another bug, Linda never inspected the whole branch when finding topology similarity- this was fixed. This bug was not specific just for bushes, but bushes as more chaotic structures then trees, were much more affected by this problem.
What we are doing right now?
We are sill working on automatic generation of special LODs, but we are very near to finishing. Today I will test the first bush with all geometry lods fully generated in Linda. We are still missing some simple post processes on 3D data, like automatic assigning of definition of penetration materials or sharp edges and so on, but this could be made manually before the final tools will be ready for full production usage. We now face the last two biggest problems with this system. First one is that the geometry models are too precise and complex, so we need to simplify the outgoing result for better ingame performance (simplify polycount of blob component). Second, sometimes there are too many components that are similar but not the same, this can increase the number of components too much, so we are working on a system, that will find similarity under some threshold and attach these components into a single one (reduce the number of components).
There are some unexpected benefits from this work in the future. We found that geometry "blobs" used for collision or view detection could be usable as volumes for light attenuation in some future lighting method. Another unexpected benefit is the use of clusters ( see attached pictures) used for making blobs on polyplanes as information about density and finding the center of mass of a branch for some future usage.
What you see on pictures?
In the first picture you see the results, color parts shows components, how they're located by Linda - ArmA 2 (right), gray picture original geometry LOD fom Armed Assault (left). See how much more precise it is, much more then the original hand made one in Armed Assault. We hope that it will resolve most of the ingame troubles, such as: "I can see but not shoot, or AI can spot me though dense vegetation...".
This picture shows clusters- density markers in the volume of a branch and uses them for convex component calculation - (bottom part of picture). On right side is fire/view geometry, on left part of picture, collision geometry.
Impact on gameplay
In our engine the AI try to spot you through the View Geometry, just a small part of your body is enough to be able to recognize and track you. When this view geometry is not masking foliage well, the chance that you will be spotted is high. Another problem is Fire Geometry, there were just trunk not foliage components, so for example, when you were hidden in some bushes and the AI saw you, looked to Fire Geometry to see if it can open fire, this geometry was empty, so they started to shoot. Empty foliage Fire Geometry also allows you to throw grenades or shoot rockets into the vegetation as if to an open space, all these things will now work better and more correctly. The reason why we do this now is that making geometry for vegetation is not a simple task. On a typical model you can use resolution LODs as reference for the volume of these special geometry LODs, but with vegetation making that from some billboards or crossing planes it's not that easy. Making correct vegetation volume manually is very time consuming, difficult and in fact impossible. There is lot of math work in the background, that allow this automatization on vegetation models.
Flora and other small eye candy
We are still working on clutter and other non-generated models of vegetation especially flora, next blog entry will hopefully be about that. But first picture with sunflowers are the results of experimental flora generating in Linda, so we are trying to also use Linda there.