The next clip shows a part of Benjamin Avery's PhD thesis, exploring the use of a head mounted display in order to view the scenery behind buildings (as long as they are brick-walled buildings). If understood correctly (and I couldn't find the relavant paper online to check this up), the overlaid image is a three-dimensional rendition of the hidden scene reconstructed from images taken by a previously positioned camera.
The interesting thing here is that a simple visual cue, such as the edges of the occluding items, can have such a dramatic effect on the perception of the augmented scene. It makes one wonder what else can be done to improve augmented reality beyond better image recognition and brute processor power. Is it possible that intentionally deteriorating the augmented image (for example, making it flicker or tainted), will make a better user experience? After all, users are used to see AR in movies, where it looks considerably low-tech (think Terminator vision) compared with what we are trying to build today.
Anyway, here's Avery himself, presenting his work and giving some more details about it: