3D mapping potential of COSMO-SkyMed sensor: definition of an image matching strategy

The topic of this paper is the presentation of an image matching technique suited to generate Digital Surface Models (DSMs) following a radargrammetric approach applied to SpotLight Synthetic Aperture Radar (SAR) stereo pairs collected by the new high amplitude resolution radar sensors COSMO-SkyMed (Italian) and TerraSAR-X (German). At present, thanks to this kind of imagery (GSD up to 1 m), the traditional radargrammetric stereo-mapping approach is likely to become a serious alternative to interferometry for the Digital Surface Models (DSMs) generation from Synthetic Aperture Radar (SAR) imagery. Nonetheless, DSMs accuracy level is strictly related to image orientation and matching strategy. A crucial issue is the definition of an image matching strategy to perform 3D surface reconstruction. Generally we can distinguish two classic techniques to carry out the matching process, the Area Base Matching (ABM) and the Feature Base Matching (FBM): these techniques do not appear suited to manage the strong geometrical deformation and the complex and noisy radiometry of SAR imagery, if separately used. Therefore, here we propose an advanced matching method based on a coarse-to-fine hierarchical solution with an effective combination of geometrical constrains and an Area Base Matching (ABM) algorithm. The suite for radargrammetric approach has been implemented in SISAR (Software per Immagini Satellitari ad Alta Risoluzione), a scientific software developed at the Geodesy and Geomatic Institute of the University of Rome “La Sapienza”. In order to investigate the 3D mapping potential of SpotLight imagery, the implemented radargrammetric tool was ap-plied to a COSMO-SkyMed stereo pair acquired over the area of Merano (Northern Italy), where a quite good LiDAR Digital Elevation Model is available as ground truth. The results are quite en-couraging, with an accuracy of about 2 meters over open areas and a quite dense points cloud even over forested areas, where it is possible to estimate the average canopy height with a mean error of about 4 meters.