Working with several researchers from the University of Washington, Microsoft's Richard Szeliski has developed a technology that converts digital images into 3D model, allowing the user to move, walk or fly into the scene.

It's called Photosynth, and it has been introduced (also) at the SIGGRAPH conference of Boston. The method consists in compiling unique features from different photographs and cross-referencing them against other images, looking for similarities.
That enables it to isolate a specific 3D position and then calculate the camera's location when the picture would have been taken.

"Basically, it's just a geometry problem. You are simultaneously adjusting the position of the camera and where those little pieces of images are, until everything snaps together"

The system can be used with only 2 pictures, but it's much more interesting when several dozen pictures are combined. Szeliski believes that cities and tourism boards, and even photo-sharing sites will soon use this technology. Flickr, Picasa, Talk a Pic ?
Here, a video is available but disappointing. It becomes an habit for Microsoft...

We take the opportunity to mention a great site. Let's imagine that you want to take a photograph of a building in the center of Beijing. Impossible to shot without a tourist or two, minimum. Shot anyway, without moving. Provide the pictures to Tourist Remover which, as indicated, is able to generate a picture with nobody.

Source: BBC News, two weeks ago.