There is a company, Chris Bell Creative Design Studios, that is developing a new technology called Virtual Earth Project, or vEarth for short.  If I understand correctly, this technology can take data from photo imagery and convert it to autogen objects which are then placed over the photo scenery.  There are some videos on their forum page, the first of which shows tree autogen.  The project is still in its infancy, and the developer emphasizes that it is being developed for FSX and P3D only.  Maybe some other enterprising genius will figure out how to do this for X-plane, once it can be demonstrated that the tech works as claimed.  One can hope, right? :smile:
 
Below is a link to the forum page.  Scroll on down past the videos to see some further discussion of the project.

http://forums.nightenvironment.com/topic/268-secret-project-unveiling-vearth-virtual-earth/

 

 

Tom

Calculating trees from photoscenery isn't actually too difficult. I have a working prototype in W2XP I did earlier this year which uses G2XPL tiles and with some input from a user about which colours the trees are, it generates trees and forests. It heavily relies on the fact that the photoscenery is colour corrected and the user has given the tool some examples of trees. It worked and makes believable scenery and it was what I was using to make my Poland scenery. The same could be extended to work out autogen zones, but generating buildings is another thing entirely. For this to work, you need very high resolution imagery. If you are actually after the individual building footprints then it's no easy task, and there are all sorts of algorithms out there to do this, from machine learning to those that work in combination with LIDAR data or point clouds. All cost $$$. For flight simulation, I guess just knowing the approximate locations of the buildings would be enough.

Its easy enough to do, using freeware tools and open source data. Its difficult to do it well though so its believable in the sim. Getting it accurate involves cross referencing multiple data sets to categorise the autogen accurately and assign fairly realistic textures.

Yep, getting the correct building types from just ortho imagery is only one part of the challenge. Luckily for the US there is actually fairly good data available (although the quality can vary from county-to-county). In Washington State for example, there is landuse data which will tell you exactly what each type of building is.