高分解能衛星画像の影解析及び反復的3Dモデリングによる建物の高さ推定

抄録

High-spatial-resolution satellite imagery over urban areas includes small and complicated shadows generated by buildings, trees and so on. These shadows affect applications such as land cover classification and traffic surveillance, but also can be used for estimating building heights, as reported by several investigators. In the present paper, we propose a novel method of estimating building heights based on the consistency between a shadow map generated from a high-resolution satellite image and that created by iterative 3D modelling with building contour data and a digital elevation model (DEM). In checking the consistency between them, tree shadows which may reduce accuracy are excluded by shadow tracking with a vegetation index map. The method was tested using a WorldView-2 image and a 5 m-resolution DEM at the Hitachi Campus of Ibaraki University (Area A), a low-rise residential area (Area B), and a low-to-mid-rise residential area (Area C) in Hitachi City, Ibaraki, Japan, and the results were compared with heights from drawings (only Area A) and laser survey data (Areas A, B, and C). In Area A, the root-mean-square (RMS) errors for 31 buildings were 1.32 m (drawings) and 1.65 m (laser survey), and those for 29 buildings without a large error were 0.78 m (drawings) and 1.18 m (laser survey). In Area B, the RMS error for 87 buildings was 1.95 m, and that for 79 buildings without a large error was 1.54 m (laser survey). In Area C, the RMS error for 99 buildings was 4.01 m, and that for 65 buildings without a large error was 1.50 m (laser survey). Major error factors in these tests were shadow disturbances caused by objects excluded in 3D modeling, such as trees and cars, and merging of shadows due to crowded buildings. Thus, the method should be applied to selected buildings but not to all buildings, but it has some robustness. In our next study we aim to predict the reliability of the estimated height, to detect errors in building GIS data, and to apply the method to foreign cities with neither building GIS data nor aerial laser survey data.