Detection and reconstruction of human scale features from high resolution interferometric SAR data

In contrast to optical imagery, modern high resolution IFSAR sensors deliver intensity images and corresponding interferometric height and coherence data from a single flight path at pixel sizes of 30 cm to 10 cm. Due to the all-weather, day-night applicability of SAR sensors, multiple views over short time can be easily obtained. However, many image users find it difficult to "read" radar images. They differ from the natural human visual impression and the familiar analogy communicated by optical imagery. The goal of our work is to convert radar data into models of the terrain and render those models in analogy to the optical sensing approach that the human visual system represents. Therefore intelligent combinations of multiple views and measurements from all IFSAR data sources available are necessary to help to overcome problems inherent in the SAR data as e.g., blur, speckle, layover and shadow. Using basic image analysis methods we present in this paper our first fully automated approach to separate buildings from other objects in an IFSAR dataset. The building shapes are reconstructed using a simple building model and the results are compared to measurements made from optical imagery.