Conventional methods for image registration tend to be restrictive, fragile, and computationally intractable, and not easily applicable to real-time video-to-reference registration. ImageCorp, Inc. proposes a novel approach to this problem, based on the present and past work of Prof. Rama Chellappa's group at the University of Maryland and at ImageCorp, Inc. First, an initial alignment is computed using platform and sensor parameters, if available. This is followed by two complementary feature-based methods, the first relying on point correspondence, and the second on global feature attributes. Methods are proposed for real-time performance self-evaluation. Phase I work will involve design of the overall system, and implementation of candidate registration algorithms. In Phase II, a prototype version of the system, capable of robust, autonomous and real-time sub-pixel registration of video to reference imagery will be developed. Potential military applications of this project are in autonomous navigation, UAVs, MAVs, UUVs, battlefield damage assessment, surveillance and security. Commercial applications include search and rescue missions, environmental monitoring, disaster relief, archeological studies and exploration etc.
Keywords: Video Sequence, Reference Imagery, Metadata, Registration, Geolocation, Feature Correspondence, Feature Tracking, Autonomous Navigation
