Abstract: Automatic identification and extraction of man-madeobjects from aerial images for cartographic purposesare currently the most challenging problems inphotogrammetry. Automating the identification andextraction procedures can significantly improve theefficiency of photogrammetric tasks, such as mapcompilation, DEM and orthophoto generation and citymodeling. This paper presents an approach for objectextraction from large-scale aerial images. The approachcomprises two key concepts: extraction is performedfrom the object view point, and objects are extractedsemiautomatically. Integration of both concepts in aconventional, well-established photogrammetricprocedure complies with the primary requirement of themapping community, which is to have an efficientprocedure, while maintaining the necessary accuracy andreliability. The paper presents the idea and themotivations, and discusses basic tools for extractingtwo types of common man-made objects: buildings androads. Preliminary results are shown and discussed aswell. !25
展开▼
