Microsoft has included the Seadragon development team as part of its Virtual Earth group in the Microsoft Live Labs. The developers became Microsoft employees on the acquisition of Seadragon in 2006. The move opens the way for much tighter integration of Seadragon into the Virtual Earth offering.
The Seadragon team developed the Photosynth product out of technology from the University of Washington. It analyses digital photographs to build a 3D point cloud of a photographed object.
There are two steps in the process. The first step is an analysis of multiple photographs of the same general area. Each photograph is scanned with a scale-invariant feature transform algorithm. It identifies features such as the corner of a window frame or a door handle and then looks for the same features in the other photographs.
By analysing the relative positions of features in each photograph, the program can define the orientation of one photo with respect to another, and then – provided there are enough photos – identify the 3D position of each feature.
In the past, photogrammetrists have attacked Photosynth for a lack of rigour. It uses computer graphics techniques rather than photogrammetry to compare pictures. On this argument, Photosynth is adequate for tourist snaps of the local church or mosque, but not for urban model building.
So far, Photosynth has not been released as a product, although it has been exposed on test sites, including the BBC site and the Microsoft site itself. The underlying Seadragon technology is also used to place Birdseye imagery on 2D orthophotos used on the maps.live.com site.
One obvious application of the technology would be in stitching together 3D street level views on Virtual Earth. However, Microsoft is concerned at the possible privacy implications of such a move.
Arch rival Google has found itself in hot water over privacy issues as the community debate heats up.
Microsoft has been experimenting with routines that automatically blank out all imagery from the street, replacing all the cars with a generic street – but it is not clear how well this can remove people nor how valuable the street information may be to potential users.