Building recognition for mobile devices: incorporating positional information with visual featuresR. Hutchings, Walterio Mayol-Cuevas, Building recognition for mobile devices: incorporating positional information with visual features. CSTR-06-017, Computer Science, University of Bristol. December 2005. PDF, 2700 Kbytes.
In this paper we describe a system that can accurately match buildings in photographs taken with mobile devices. Computer vision methods are enhanced by a novel use of the GPS (Global Positioning System) position of a user. With a query image s GPS location, an approximation of the view of each building from this position in world space can be calculated. Buildings are represented as planar surfaces and an efficient geographical database positions each building in world space. By using planar rectification of the building image, the effects of perspective change are accounted for. With such planar representations, accurate geometric analysis can then be used to help determine the correct building match. With these methods, a recognition system has been formed that has been shown to be accurate for a wide range of viewpoints and image scaling of a building.