They fly a plane over the terrain in some kind of raster flight path. The plane has a number of camera on board. The number was said to be secret but I could believe it to be a lot - like 16 or something like that. They have been bolted down with fairly accurate position and such that they overlap. They did say they were digital SLR cameras.
So they know the planes position very accurately - probably to about a few cm's. They know the flight path at the time to a similar accuracy and they have the camera positions and directions known. Then they match up multiple photos from different shots at different times. Because you know where you were at the two times you can see the differences that show up on the spacing between features across two or more shots. This gives you Z depth. Working back from that plus the planes position can give the XYZ position of the ground terrain. Normally you still get shadow zones from the parts of the photos which were obscured but if your taking many many shots as part of a raster flight path then there is going to be very little left obscured. Basically only stuff underneath a cover.
You can do some of the math and processing to scan in a *sparse* point cloud easily using your own digital cam and some free software. I have wrote a blog about that - you might need a login to see pics, not sure:
http://www.cartographersguild.com/blog.php?1311-Redrobes-blog-about-playing-with-the-technical-side-of-graphics.
These guys tho are getting more than a sparse point cloud out of it and are post processing it a bit more to generate the buildings as vectored meshes. I can see from the video that they go beyond this to work out which meshes are buildings and which are roads or terrain etc. They are classifying the mesh output as well. So they are well beyond the normal consumer state. But then they are military and well funded...
I didn't realize Nokia have them. That's quite cool. All the demos were on iPads
