Object-based multi-view façade matching in SAR images of dense urban areas

Abstract

Building information retrieval from remote sensing imagery is an important research topic in the remote sensing field. The extracted building footprints, heights and façade structures help us better understand our living areas. This information also gives great assistance for large urban area monitoring and future planning. With the availability of meter-resolution spaceborne synthetic aperture radar (SAR) sensors such as TerraSAR-X, SAR data have also been extensively exploited for building 3-D reconstruction. On one hand, SAR interferometric (InSAR) methods use the phase information of a time series of observations to retrieve the topography of urban areas. InSAR methods can provide 3-D information with detail up to individual floors, given a relative large stack of images. On the other hand, another group of methods extracts building information by solely exploiting the amplitude of a few SAR images, e.g. stereo radargrammetry. These methods require fewer images than InSAR methods. But the retrieved 3-D information is so far only limited to a single building roof height or the height of a set of sparse points in the image. Therefore, this thesis proposed a new method that provides precise 3-D height estimates of individual floors of a building façade using only three SAR amplitude images of different incidence angles from either all ascending or all descending orbits. The firsthand information of a building can be obtained within a short time span. The method detects and extracts building façade with regular patterns, and performs an object-based façade matching and floor height estimation. The proposed method requires no coregistration between the three images. This avoids additional coregistration error, as well as the azimuth disparity introduced by coregistering images acquired from nonparallel orbits. The algorithm has been tested on real and simulated data. The object-based façade matching and height estimation is robust against the appearance of strong reflective scatterers that do not belong to the building façade. In addition, the algorithm is applied to individual buildings whose heights of each floor are referred to their ground floor respectively, so that the effect of atmospheric delay is insignificant.