| This dissertation presents a suite of algorithms for the automatic analysis and understanding of images derived from a high resolution airborne Synthetic Aperture Radar (SAR), with the specific goal of constructing two- and three-dimensional site models. A site model is a delineation of man-made structures and objects (buildings, roads, vehicles), as well as natural terrain features (vegetation, lakes). If only a single SAR image is available, the site model is essentially a terrain map of the scene. Heights of buildings and vegetation canopy are extracted from multi-pass imagery, which are registered using a simple transformation derived from the sensor acquisition parameters. Interferometric SAR measurements when available, provide elevation data for the imaged area, which is then used to detect and extract building heights to construct a three-dimensional site model. Application of site models for context-based image exploitation are demonstrated for false alarm reduction in automatic target recognition and wide-area surveillance applications. |
