ARUBA and TOBAGO { new approaches for the generation of Cybercity from images

Automated methods for reliable 3-D reconstruction of man-made objects are essential to many users and providers of 3-D city data, including urban planners, architects, telecommunication and environmental engineers. Manual 3-D processing of aerial images is time consuming and requires the expertise of qualiied personnel. Therefore, the necessity to interpret, classify and quantitatively process aerial images in a semi-or fully automatic mode and to integrate the results into CAD-or spatial information systems is more urgent than ever. This paper describes current research activities at ETH Z urich in automated 3-D reconstruction of buildings. Two conceptually diierent approaches to building reconstruction are presented: TOBAGO and ARUBA. The rst approach is semi-automatic and aims at the automatic structuring of manually measured 3-D point clouds to generate CAD models of complete buildings. It is geared towards a reliable and exible procedure which has a large potential to be useful for professional practice. The second approach aims at fully automatic detection and reconstruction of buildings. This automatic system relies on hierarchical hypothesis generation in both 2-D and 3-D. To achieve this we have developed novel methods for feature extraction, segment stereo matching, 2-D and 3-D grouping, color and object modeling, and geometric reasoning. We also present and evaluate the results of both methods on various data sets. We will also show how modern visualization tools can be used for 3-D animations ((y-overs and walk-throughs). For that the vector data set will be augmented with image raster data from still video images of the facades to generate photo-realistic representations.