Building detection and building parameter retrieval in InSAR phase images
Building detection and building parameter retrieval in InSAR phase images
复制标题
DOI：
10.1016/j.isprsjprs.2016.02.009
复制
发表时间：
2016-04
影响因子：
12.7
通讯作者：
C. Dubois;A. Thiele;S. Hinz
中科院分区：
文献类型：
--
作者：
C. Dubois;A. Thiele;S. Hinz
目前卫星SAR任务提供的高分辨率使其成为对城市地区进行详细分析的有吸引力的解决方案。特别是由于它们与天气和日光无关，当光学传感器达到极限时，它们可以被使用。由于这种SAR传感器的特殊斜侧面配置，在建筑物位置会出现停留、双反弹和阴影等现象，这些现象可以通过非常高分辨率（VHR） SAR数据更好地理解。检测这些区域，以及通过对提取的结构进行详细分析来检索建筑参数，是一项具有挑战性的任务。事实上，根据采集配置、建筑材料和周围环境的不同，这些模式在振幅SAR图像中并不总是一致的。它们很难被自动识别和区分。考虑InSAR相位图像而不是振幅图像对这项任务非常有帮助，因为InSAR更依赖于几何形状。因此，本文主要研究InSAR相位图像中建筑物停靠物的检测与提取。提出了两个互补的探测器，并将它们的结果进行了组合，以提供可靠的构建假设。根据提取的片段进行进一步分析。特别分析了连接立面的数量。最后，每个立面都采用了独特的几何形状，以确定最终的建筑参数，如长度、宽度和高度。该方法的结果显示了三个不同的数据集，首先是在提取的正确性和完整性方面，其次是在提取的建筑参数的准确性方面。对于考虑的数据集，完整性和正确性分别约为70%和90%。排除明显的异常值，确定的参数的精度可达4米（长度），2米（高度）和3°（方向）。本文考虑的是孤立的、中高层的平顶矩形建筑。
The high resolution provided by the current satellite SAR missions makes them an attractive solution for the detailed analysis of urban areas. Especially due to their weather and daylight independency, they can be employed when optical sensors come to their limits. Due to the specific oblique side-looking configuration of such SAR sensors, phenomena such as layover, double bounce and shadow appear at building location, which can be better understood with very high resolution (VHR) SAR data. The detection of those areas, as well as the retrieval of building parameters through a detailed analysis of the extracted structures, is a challenging task. Indeed, depending on the acquisition configuration, on building material and surroundings, those patterns are not always consistent in amplitude SAR images. They can be difficult to recognize and distinguish automatically. Considering InSAR phase images instead of amplitude images is very helpful for this task, as InSAR is more depending on the geometry. Therefore, in this paper, we focus on the detection and extraction of building layover in InSAR phase images. Two complementing detectors are proposed, and their results are combined, in order to provide reliable building hypotheses. Based on the extracted segments, further analysis is conducted. Especially, the number of connected facades is analyzed. Characteristically geometrical shapes are finally fitted for each facade to permit the determination of the final building parameters as length, width, and height. Results of this approach are shown for three different datasets, first in terms of correctness and completeness of the extraction, and second in terms of accuracy of the extracted building parameters. For the considered datasets, the completeness and correctness are of about 70% and 90%, respectively. Eliminating clear outliers, the determined parameters present an accuracy up to 4 m (length), 2 m (height) and 3°(orientation). In this article isolated, middle to high rise buildings with flat roof and rectangular shape are considered.