期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2007
卷号：XXXVI-3/W49B
页码：13-18
出版社：Copernicus Publications
摘要：The need to automatically extract topographic objects, especially buildings, from digital aerial imagery or laser range data remains an important research priority in both photogrammetric and computer vision communities. This paper describes the proposed model for Level-of-Detail building modelling and progress with the prototype implementation. The paper begins with an overview of the concept of Level-of-Detail, important for adaptive building modelling. Building regions of interest are derived from a normalised digital surface model (nDSM) and regularisation of the roof lines is achieved by a set of contextual constraints with particular emphasis on rectangular buildings. For detailed building reconstruction, the main consideration is given to polyhedral building types with limited support for curvilinear shapes. A moving least squares approach for computation of surface normal vectors and texture metrics is employed for planar segmentation of both gridded data and unstructured point clouds. Delineation of homogeneous planar segments is based on a distance metric between neighbouring local planes. 2-D edge lines derived from the orthoimage are matched with 3-D lines derived from LiDAR based on adjacent plane intersections and then used for the final building reconstruction. Connected regions which fail the local planarity tests and are sufficiently large, are segmented using curvature measures based on least squares quadric surface fitting. Provisional results from the algorithms are promising
