Object extraction from Remote Sensing images and Image / Elevation data aggregation for Realistic 3D Rendering

We present a complete processing line to enhance large Remote Sensing (RS) image databases in the framework of real-time 3D rendering. RS and Virtual Reality communities are faced to several challenges to deal in real-time with large image datasets while preserving high quality and realistic rendering. There is an increasing interest of InSAR (Interferometric Synthetic Aperture Radar) data, due to principally with the availability of a near global coverage of HR DEM with the Shuttle Radar Topography Mission (SRTM). Important improvements of DEM are needed to provide more realistic visualizations. DEM pre-processing are performed by exploiting only the couple of data required for 3D visualization. A non stationary bayesian filter is presented to remove noise and small artifacts which pervade the InSAR DEM while preserving structural information. To produce more realistic rendering, information from optical images are fused in the DEM data. An object based description of large optical Remote Sensing (RS) images is obtained in two stages. (i) The optical image is segmented to simplify the amount of data. (ii) A novel dynamical algorithm is proposed to extract the regions and geometry which are stored in a tree structure that defines spatial relationships between regions (e.g. adjacencies, inclusions). An interactive object selection procedure links the extracted objects to their elevation modeling. The objects 3D structure is estimated from the DEM data using its planimetric shape extracted from the optical image and the elevation modeling assigned. The proposed object line processing provides more realistic 3D visualizations.