Less than 24 hours is all your need to build Rome these days: a team of developers from the University of Washington and Cornell University has come up with an algorithm that can aggregate thousands of tourist photos from social network photo-sharing websites and create a three-dimensional virtual city model from them.
Highly popular tourist sites such as Rome work well currently there are more than two million photos of Rome on Flickr. The Washington University team has also used its technology to recreate the cities of Venice and Dubrovnik.
But how does the technology work? And how can the work be done so quickly? Noah Snavely, assistant professor in the Computer Science Department at Cornell University, told me that there have been a couple of innovations. He said: “Possibly the main one is that we figured out how to exploit large clusters of machines, which has enabled us to then work with large collections of images.”
The algorithm developed by the team is able to run simultaneously on many different machines including remote servers through an Internet connection which multiplies the power and speed of the calculation significantly and has enabled the team to attempt more complex projects.
The second innovation is that the newly developed computer algorithm is able to establish ‘matches’ for the photos. Snavely explained: “If we have 200,000 images of a city, we don’t know which ones ‘match’ each other – some of the photos might be irrelevant, so figuring out which sets of photos go together is important.”
A 3D image using a type of ‘point cloud’ system can then be created (see this article for how 3D scanners created point clouds and a virtual model at the catacombs in Rome). Even the position of the photographer at the time the photo was taken can be calculated.
Sameer Agarwal, computer science and engineering assistant professor at the University of Washington, says that previously this process would have taken a very long time. He told physorg: Even if we had all the hardware we could get our hands on and then some, a reconstruction using this many photos would take forever.”
Previous attempts to create virtual 3D models using photo-stitching techniques have had limited success. Programmes such as Photo Tourism and Photosynth have been able to recreate individual landmarks but the results obtained with Photosynth seem fairly patchy, as this virtual view of the Pantheon in Rome suggests (sorry but it made me feel a bit seasick!).
The Washington team downloaded 150,000 photos of Rome from Flickr and in 21 hours they were able to create the 3D digital model which enables a viewer to ‘fly’ around the city’s main sites, including the Pantheon and the Colosseum.
There is of course huge potential for this technique and Snavely is optimistic that the algorithm will be used by archaeologists in future. He said: “The plan is to release the code and put it in a form that anyone can use. We are planning to apply it to other large cities, such as Manhattan and London.” He adds that he is very interested in heritage work and is already working with an archaeologist to develop the use of the code for heritage sites.
The idea of creating a virtual Venice for people to visit instead of the actual sinking version of the city has already been suggested. Other possible uses include virtual online maps, automatically creating cities for video games, and digitally preserving certain cities. The possibilities for virtual viewing seem almost endless.
I’ll be speaking to the brains behind the operation very soon for a Heritage Key interview, so keep an eye on this page to find out what the team have planned next. You can also get vitual with Heritage Key’s Google-Earth-based 3D Rome here.