With a set of photos of the same static scene, it's possible to figure out the camera position, angle, and focal length of every camera, sometimes to millimeter accuracy. You just have to correlate the same features in each frame, then solve it.

3d motion trackers do this. (Blender.org has one. You can solve this same problem using it. Bring in public and street view photos of the area. It's a bit of manual work to put all the images in a movie then go frame by frame and tag features... But you could get much more accurate than the article.)

The power is in having access to large databases of photos of everywhere, and even better if all the photos were taken the same way, like google's street view... and large amounts of computing resources to index and correlate all of the features from all of the images automatically.

I would be surprised if there is not a project in some huge company or lab that can identify features in nearly any public outdoor shot and determine exactly where it was taken. (perhaps for intelligence or forensics)

I believe this paper introduced me to the idea in 2006: http://research.microsoft.com/en-us/um/redmond/groups/ivm/ph...