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：2003
卷号：XXXIV-3/W8
出版社：Copernicus Publications
摘要：The performance of a system for semi-automatic building extraction from digital aerial images is evaluated. Using this system, building s are reconstructed by a combination of parametric primitives that are approximately positioned in the images by a human operator , whereas their parameters can be determined precisely by automatic procedures. Unlike in similar systems, the parameterisation of th e primitives can be changed in the process of reconstruction to adapt them to irregular building shapes, which is made possible by a specific method of representation of the primitives in parameter estimation. A test project using images of different scales and overlap for the reconstruction of a densely built-up area was carried out. Depending on the image scales, the buildings were reconstructed at th e best possible levels of detail. The time required for reconstruction and the success rates of automatic fine measurement were assessed . For distinct buildings, the accuracy of automatic fine measurement was determined, showing that if the automated module is successful , it gives very accurate results. The project showed the benefits of primitive adaptations for the reconstruction of irregular buildings i f a high level of detail is to be achieved, whereas at lower levels of detail, the homogenisation by using regular building primitive s comes to the fore. Up to 87% of the most salient building parts could be determined by automatic fine measurement, the success rat e improving with the number of overlapping images. The time required for building reconstruction was found to be in.uenced heavily by the interpretability of the images
关键词：Building Extraction; Semi-automatic; 3D City Models; Model-driven Matching
