Georegistration of multiple-camera wide area motion imagery

Abstract

Wide area motion imagery sensors utilize multiple cameras on a single aerial platform to monitor very large geographic areas in real time. The images must be stabilized and georegistered before they can be combined with other geospatial datasets, but their wide fields of view and oblique viewing angles make it difficult to align them accurately with a geographic reference frame. We describe a georegistration algorithm that accepts a digital elevation model as a geographic reference from which it generates predicted images. Registration of these predicted images produces 3D-to-2D tie points that determine the motion imagery camera models. We present results on multi-camera motion imagery from the U. S. Air Force's CLIF 2006 dataset using a bare-earth U. S. Geological Survey digital elevation model. The algorithm accurately georegisters the imagery despite the lack of buildings and trees in the model. Because of the wide availability of digital elevation models, the algorithm provides a practical means of georegistration.