Research On Three-Dimensional Reconstruction Of Buildings From High-Resolution SAR Data

The three-dimensional (3D) reconstruction of buildings is a technology to obtain the spatial 3D structure information of buildings from SAR data/images. It is significant in many remote sensing applications, such as urban development and planning, dynamic sequential monitoring of cities, infrastructural damage and economic losses assessment after disasters, urban moving target research, etc. This dissertation is based on the metric and submetric high-resolution synthetic aperture radar (SAR) data, and takes the common buildings in practice with basic cuboids and combinational structures as the research objects. According to the typical electromagnetic scattering features of these buildings in SAR images, the 3D reconstruction of their main structures are realized by the exact extraction of scattering features and the accurate estimation of geometric parameters. The main works and contributions of this dissertation are presented as follows:Firstly, the SAR imaging mechanism and main scattering features of buildings are researched. The mapping relationships between 3D structures and SAR image appearances of buildings are systematically summarized. The imaging characteristics and their influence factors are then discussed. Afterwards, the formation reasons, imaging characteristics and influence factors of the main scattering features of buildings in SAR images, i.e., the layover, double bounce and shadow, are analyzed in detail.Secondly, the SAR scattering feature extraction technique for buildings is studied. Aiming at the high sidelobe interference of point spread function in the traditional extraction methods, we start from echo domain and propose the sparse reconstruction technique combining Hough transform detection for double bounce, the sparse reconstruction technique combining TV constraint for strong backscattering, and the sparse reconstruction technique combining amplitude reversal and TV constraint for shadow, respectively. These methods reduce the sidelobe interference of point spread function and improve the accuracy in the scattering feature extraction. Simulation results and method comparisons show the effectivenesses of the three methods.Thirdly, the geometric parameter estimation technique for buildings is researched. For the case when the SAR data condition is good,.a parameter estimation method based on solid geometry inversion is proposed, which has better estimation accuracy and computational efficiency. For the case when the SAR data condition is not ideal, an alternate iteration and mutual information matching method is proposed, which can accurately estimate multiple geometric parameters with a good balance between performance and computational complexity. Simulation results and method comparisons demonstrate the effectivenesses of these two methods.Fourthly, the 3D reconstruction technique for combinational buildings is studied. The existing 3D reconstruction techniques which mainly focus on buildings with basic cuboid structures are further developed to more complex and common buildings with combinational structures, and a structure filling and matching method is proposed. We construct the structure model of combinational buildings by extracting, locating, labeling, refining and filling their "structure elements". Based on this, the structure matching is performed via the combination of mutual information matching and multi-parameter optimization. Then the accurate 3D recomstruction of combinational buildings is completed. The performance of the proposed method is evaluated on TerraSAR-X data and real digital elevation model (DEM) data to show its effectiveness, reliability and accuracy.Fifthly, the fusion and reconstruction technique for buildings under multi-aspect SAR data is researched. To solve the problems of high false alarm and high sidelobe interference in building footprint extraction from single-aspect SAR data, a "sparse reconstruction-fusion-confidence score test" method for footprint extraction using multi-aspect SAR data is proposed. To solve the problems of backscattering aliasing, shadowing and unknown of structure in 3D reconstruction of building in complex scene, a structure modeling and fusion method is proposed. Simulation analysis and measured data verify the effectivenesses and reliabilities of the two method when dealing with their problems, respectively.