Object based fusion of SAR-data with oblique aerial imagery in urban areas (2016)

Man-made structures like urban areas are often characterized by simple and repeated geometric forms. This leads to preferred rectangular and regular alignment of objects like windows or balconies at façades of the majority of buildings in modern cities. These repetitive patterns are also detectable in Synthetic Aperture Radar (SAR) acquisitions of the scene. Modern SAR systems like TerraSAR-X or Cosmo-Skymed are weather independent and offer geometric resolutions of up to 60 cm in the sensing coordinate system. If a time series of SAR acquisitions is available the Persistent Scatterer Interferometry (PSI) technique allows obtaining distinct points, i.e. Persistent Scatterers (PS), corresponding to temporally stable and strong scatterers compared to their surroundings. Furthermore, the velocity of such points can be estimated with a precision of some mm/year. Since the trihedral corners of windows or balconies induce those strong reflections, PSI allows to monitor building movements due to subsidence with a high accuracy (some mm/year) from space.

Another well established remote sensing data is optical oblique imagery. It provides much higher geometric resolution and easier interpretability for human operators. The integrated use of both data types in a joint data interpretation framework would allow for many new applications in the field of change detection. Tasks like time critical change detection after natural disasters, for instance, demand this very high geometric resolution but also a weather independent data acquisition immediately after the event. Another conceivable scenario would be if a stack of SAR acquisitions constitutes the pre-event situation while oblique aerial imagery captures the situation afterwards.

The objects of interested are mostly urban built-up areas which show a drastically different appearance in both types of data due to the different imaging geometry and wavelengths. Instead of looking for single window correspondences, we try to establish the link between both data domains by the salient regularity which is preserved in both imaging geometries. Figure 1 shows the appearance of three facades in an oblique optical image (a) and in a corresponding SAR acquisition (b).

Figure 1: Appearance of facades in an oblique optical image (a) and a SAR acquisition (b). Corresponding facades are marked in the same color. The regular structures are present in both data domains.

The research is focused on finding this regularity and establishing a correspondence between both data domains. The regularity is expressed as a lattice consisting of Persistent Scatterer in the SAR domain and repetitive image tiles for the optical imagery, respectively. Since the data appearances of the same façade lattices are different due to the used wavelength and imaging geometries, the definition of correspondences is mainly restricted to a topological matching. Geometry information has also to be introduced to regard the anisotropic error contributions of PS.