We present a technique to extract complex suburban roofs from sets of aerial images. Because we combine 2-D edge information, photometric and chromatic attributes and 3-D information, we can deal with complex houses. Neither do we assume the roofs to be flat or rectilinear nor do we require parameterized building models. From only one image, 2-D edges and their corresponding attributes and relations are extracted. Using a segment stereo matching based on all available images, the 3-D location of these edges are computed. The 3-D segments are then grouped into planes and 2-D enclosures are extracted, thereby allowing to infer adjoining 3-D patches describing roofs of houses. To achieve this, we have developed a hierarchical procedure that effectively pools the information while keeping the combinatorics under control. Of particular importance is the tight coupling of 2-D and 3-D analysis.

We present a technique to extract complex perceptual structures for which no specific template exists. We show how to relate and group contours based on similarity in position, orientation, and photometric and chromatic properties. By fusing all these sources of information we can robustly group contours on other grounds than geometric regularity while keeping the combinatorics under control. In addition we use a data-driven selection technique that drastically reduces the complexity of the task. We present the approach and show some results.

Automation of Digital Terrain Model Generation and Man-Made Object Extraction from Aerial Images (AMOBE) is a joint project between the Institute of Geodesy and Photogrammetry (IGP) and the Institute of Communications Technology (Image Science Group) (IKT) at the Swiss Federal Institute of Technology in Zurich. In the project we develop methods and algorithms to detect and reconstruct man-made objects, such as buildings and roads, and to generate Digital Surface Models (DSMs) from high resolution aerial images. Primary attention in AMOBE focuses on high quality reconstruction of buildings as being one of the more predominantly and frequently occurring 3-D man-made objects in high-resolution aerial imagery. In this paper we present our research strategy, current results, and make an outlook onto future work.

. Several methods have been proposed to reconstruct houses from aerial images, most of them using rather specific models. These methods can not be applied when dealing with european-style houses and when a high level of detail is required. We present a system that works with rules instead of parameterizable shape models. This allows a larger variability of the objects to be extracted but also results in a more complex structure of the knowledge base and in a more sophisticated reasoning control. In the application section we demonstrate the use of such a system for reconstructing an object made up of planar surfaces from an aerial image. We will also explain how additional domain specific knowledge could speed up the extraction process and increase robustness. 1 Introduction The development of practical algorithms that support the robust and precise reconstruction of buildings and other man-made objects from aerial imagery is one of the aims of the project AMOBE 1 , a join...

We present a system for the reconstruction of houses from aerial images. Most of the methods proposed sofar use rather specific models. These models do not hold when dealing with european-style houses. Our system works with rules instead of models merely based on shape. This enables us to include additional domain specific knowledge, allowing a larger variability of the objects. However, this also necessitates a more complex structure of the knowledge base and a more sophisticated reasoning control. We explain the practical application of such a system for the reconstruction of a house from an aerial image. 1 1. Introduction The acquisition of 3D descriptions of houses is currently an issue of high importance to users of geoinformation. When using aerial imagery, which has proven to be a valuable source for this task, the reconstruction of a building means to find or hypothesize all parts of the roof because from all building parts they are the best visible in the images. The intend...

Many approaches to image understanding and object recognition employ either a single model or more than one but only one for each level of processing and processing takes place unidirectionally either bottom-up or top-down, thus restricting the flow of information between models of neighboring level to a one-way. These approaches have their limits when dealing with objects that come in a large variety of appearances and shapes and, hence traditional appearance- or shape-based models can not be used. We will show that it is necessary to integrate such models with more abstract ones into one coherent system where the control implements a hypothesize-and-test cycle using all available models. Such a system has been built and its application is demonstrated on the reconstruction of building roofs from aerial images. It is based on rules and the general framework of a blackboard. Integrating Models for Object Reconstruction 3 1 Introduction Computer vision tries to answer questi...