Video-to-reference image indexing

This work is concerned with registration of data in geospatial databases, especially with registering images taken by different sensors and from different viewpoints of the same scene. This research has many applications in creating and updating maps, surveys and other geospatial data sources. While considerable work has been performed in image registration [2], extant approaches break down as viewpoint and/or sensor vary beyond relatively small changes. Further, extant image-based registration technology has only been demonstrated to support video-to-reference image alignment when initialized to within several hundred pixels of the correct result [6]. Due to errors, drop outs and otherwise limited availability of telemetry, ineffective a priori knowledge of relative video/reference image alignment is a real-world problem. The developed approach allows video-based image descriptors to index directly into a reference image database. Successful indexing into the database implies that approximate position of the video relative to the reference has been recovered; this information can then serve to initialize extant technology for video-to-reference registration. The approach provides uniform representation of video and reference imagery and a corresponding method for quantifying the goodness of match between two image samples. The representation combines image appearance, characterized in terms of texture defined regions, and image geometry, characterized in terms of spatial relationships between textured regions. By construction, the matching method is robust to a range of photometric and geometric distortions between image sources, including changes in grey-level contrast and affine geometric transformations. The developed approach has been algorithmically specified and instantiated in software. Empirical evaluations with a reference image database derived from orthoimages of distinct geographic locations that is indexed via synthetic aerial video document the promise of the approach.