3D space intersection features extraction from Synthetic Aperture Radar images

The main purpose of this Thesis is to develop new theoretical models in order to extend the capabilities of SAR images space intersection techniques to generate three dimensional information. Furthermore, the study aims at acquiring new knowledge on SAR image interpretation through the three dimensional comprehension of the scene. The proposed methodologies allow to extend the known radargrammetric applications to vector data generation, exploiting SAR images acquired with every possible geometries. The considered geometries are points, circles, cylinders and lines. The study assesses the estimation accuracy of the features in terms of absolute and relative position and dimensions, analyzing the nowadays operational SAR sensors with a special focus on the national COSMO-SkyMed system. The proposed approach is original as it does not require the direct matching between homologous points of different images, which is a necessary step for the classical radargrammetric techniques; points belonging to the same feature, circular or linear, recognized in different images, are matched through specific models in order to estimate the dimensions and the location of the feature itself. This approach is robust with respect to the variation of the viewing angle of the input images and allows to better exploit archive data, acquired with diverse viewing geometries. The obtained results confirm the validity of the proposed theoretical approach and enable important applicative developments, especially in the Defence domain: (i) introducing original three dimensional measurement tools to support visual image interpretation; (ii) performing an advanced modelling of building counting only on SAR images; (iii) exploiting SAR images as a source for geospatial information and data; (iv) producing geospatial reference information, such as Ground Control Point, without any need for survey on the ground.