TOBAGO - A SEMIAUTOMATED APPROACH FOR THE GENERATION OF 3-D BUILDING MODELS

3-D city models are in increasing demand for a great number of applica tions. Photogrammetry is a relevant technology that can provide an abu ndance of geometric, topologic and semantic information concerning the se models. The pressure to generate a large amount of data with high d egree of accuracy and completeness poses a great challenge to photogra mmetry. The development of automated and semi-automated methods for th e generation of those data sets is therefore a key issue in photogramm etric research. We present in this article a strategy and methodology for an efficient generation of even fairly complex building models. Wi thin this concept we request the operator to measure the house roofs f rom a stereomodel in form of an unstructured point cloud. According to our experience this can be done very quickly. Even a non-experienced operator can measure several hundred roofs or roof units per day. In a second step we fit generic building models fully automatically to the se point clouds. The structure information is inherently included in t hese building models. In such a way geometric, topologic and even sema ntic data can be handed over to a CAD-system, in our case AutoCad, for further visualization and manipulation. The structuring is achieved i n three steps. In a first step a classifier is initiated which recogni zes the class of houses a particular roof point cloud belongs to. This recognition step is primarily based on the analysis of the number of ridge points. In the second and third steps the concrete topological r elations between roof points are investigated and generic building mod els are fitted to the point clouds. Based on the technique of constrai nt-based reasoning mio geometrical parsers are solving this problem. W e have tested the methodology under a variety of different conditions in several pilot projects. The results will indicate the good performa nce of our approach. In addition we will demonstrate how the results c an be used for visualization (texture mapping) and animation (walk-thr oughs and fly-overs). (C) 1998 Elsevier Science B.V. All rights reserv ed.