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：2004
卷号：XXXVI-8/W2
出版社：Copernicus Publications
摘要：Airborne laserscanning become a common surveying tool in the last years. A lot of applications are based on digital terrain models in urban planning, forestry, topographic mapping, environmental monitoring or disaster management. The filtering process and the subsequent DTM generation using airborne laserscanning data can be significantly improved by classification of non-terrain objects (e.g. vegetation, buildings etc.). For this reason it is very important to classify the objects on the surface of the earth. The commonly used pixel-wise image classification methods are limited in terms of reliability of its results, especially when using laserscanning data. Therefore, a segment based classification method has been developed. The segments can be buildings, building parts, connected vegetation, or parts of it. This kind of segment based classification can be the premise of 3D object reconstruction. In the first step of object segmentation, a normalized digital surface model (nDSM) is generated by extracting ground points exclusively. The full automated DTM generation can not result a perfect DTM, some kind of sharp terrain edges and their surrounding areas can be filtered out as well. Besides the man-made objetcs and vegetation, this nDSM also contains some terrain parts. After classification, these terrain objects can be reintegrated to the data of the terrain model, and a new one can be generated in higher quality. After segmentation, different kind of object-oriented features are calculated for each segment, like height texture, first/last pulse differences, etc. A fuzzy logic approach is used to obtain a reliable building, vegetation and terrain classification based on these features. This classification is based on segment objects, so not segmented objects can't be classified. The segmentation is based on the last pulse laser data in order to avoid mixed segments (e.g. building with trees). On the other hand, as a disadvantage, not all vegetation objects are segmented, since the last pulse data set contain buildings, and partly vegetation. Therefore, a great amount of vegetation can't be classified. To solve the problem, a hierachical approach of segmentation and classification has been developed. Buildings are classified from the last pulse data, than –after masking building objects- vegetation from first/last pulse data
