3D EO/SAR Reconstructions from Single and Limited Viewing Perspectives

Scene exploitation using synthetic aperture radar and electro optical sensors is vital in supporting combat identification (CID) and intelligence, surveillance, and reconnaissance (ISR) missions. As the number of distinct sensors deployed in increases, techniques to fuse information across sensors and time are required. One promising technique to support multi-sensor multi-temporal fusion is to map sensor information directly onto three dimensional (3D) representations of the targets of interest. Projecting unique imaging geometries and modalities into a common cartesian 3D coordinate system provides a natural basis to integrate information to realize mission goals (detection, association, classification, pose estimation, change detection, etc.). Most reconstruction pipelines are reliant on having many camera viewpoints often by an airborne platform performing a circular flight plan around an object of interest. These requirements are not feasible in a military operational scenario in which air superiority is not maintained. We directly address warfighter and intelligence analyst need by developing algorithms to estimating 3D shape from a sparse set of images utilizing generative adversarial networks, structure from motion, multi-view stereo, and robust interferometry.