The University of Washington, which licensed to Microsoft the photo-stitching technology used in the Photosynth 3D photo-browsing program, has come up with a revamped algorithm that can be used to model not just individual landmarks but now entire cities using tens or even hundreds of thousands of photos.

For example, it has been used to create a digital version of Rome using 150,000 photos posted by tourists on Flickr. The model was assembled in 21 hours -- compared with the months it would have taken using the previously photo-stitching technology. The new algorithm uses a more efficient approach to match the photos, and takes advantage of parallel processing to execute the program across many computers at once.

The model of the Colosseum, in the video above, was constructed from 2,106 images. This particular video shows the overall view, but users would be able to zoom into individual images, as well.

The basic idea is to position two-dimensional pictures on screen in the places and angles that they would appear in the real world, letting people zoom around the virtual 3D scene. According to the UW, the new algorithm could see a variety of applications in the short term -- in video games, mapping programs or a new Photosynth version, for example.

Other cities that have been similarly reconstructed are Dubrovnik, Croatia (60,000 images in less than 23 hours, using a 350-computer cluster) and Venice, Italy (250,000 images in 65 hours, 500 computers).

A paper on the new technology is scheduled to be presented in October at the International Conference on Computer Vision in Kyoto, Japan. The lead author is Sameer Agarwal, a UW acting assistant professor of computer science and engineering. Co-authors are Cornell University Noah Snavely, who developed the original "Photo Tourism" project as a doctoral student at the UW; Rick Szeliski of Microsoft Research; UW computer science professor Steve Seitz; and UW graduate student Ian Simon.

For more technical detail, this piece by UW engineering writer Hannah Hickey is a great read. This page provides more information on the overall project.

Microsoft Research, Google, the National Science Foundation and the Office of Naval Research supported the research.