Building Height Extraction From High Resolution SAR And Auxiliary Optical Image

Height information extraction of man-made instruments such as buildings via remote sensing is an important issue in the area of urban remote sensing, which is of great significance for urban disaster risk warning and battle damage assessment. As two typical heterogeneous images, SAR and optical remote sensing images show strong complementariness in target information characterization. Comprehensive utilization of these two kinds of images can improve the accuracy and reliability of target interpretation. By taking advantage of optical images in planar structure characterizing as well as the SAR images in containing height information and keeping present situation, this paper conducts the research of building height information extraction from high resolution SAR images with the aid of optical data.Detection building from remote sensing images is the prerequisite for building height extraction. Based on the analysis of the building characteristics in high resolution optical images, this paper presents a building extraction method by attribute filtering and context analysis. Firstly, the attribute morphological transform is implemented on the optical image for candidate building detection; secondly, the context between building and its shadow is considered for the confirmation of buildings from the building candidates, which reduces the false alarms effectively.The presence of speckle reduces the intelligibility of SAR image. As an important preprocessing step for building height extraction, a confidence interval and morphological reconstruction based adaptive windowing method is proposed for speckle reduction, which can efficiently suppress speckle noise in homogeneous region and meanwhile maintain the structure information.Correct understanding of building backscattering mechanisms is crucial for building height extraction. By analyzing the dominating backscattering mechanisms as well as the imaging geometry for the typical flat roof and gabled roof buildings, an orthogonal projection model is established for building structure calculation in the SAR image. Based on this, the height estimation methods for isolated and partially occluded building cases are investigated.For the isolated building case, a model-based structure prediction method is proposed for height extraction from SAR images. The structure corresponding to a certain building geometric model hypothesis is predicted by the orthogonal projection model and then mapped onto the real SAR image for matching, which transformed the height information extraction into the problem of finding the maxima of the matching function. Compared with the existing method, the proposed method has the advantage that neither SAR image simulation nor feature extraction is required.Based on the analysis of the building structure distortion caused by the occlusion effects, an iterative model matching method is proposed for partially occluded building height extraction. The exact visible facets of the mutual occlusion buildings along the radar line-of-sight are identified and then mapped onto the slant range plane following the orthogonal projection model for occluded building structure prediction. Meanwhile, the genetic algorithm is also adopted to accelerate the high dimension parameters optimization of the likelihood function. The experimental results with the simulated data sets and real SAR images show that the proposed method could efficiently estimate building height from SAR imagery with the aid of optical image data, and achieves reliable results with the partial occlusion case.