uu
Jherence of
08/05.
ion using
sed on intensity
ledge.
ion using
sed on inten-
Formation and
e.
PROJECT AMOBE: STRATEGIES, CURRENT STATUS AND FUTURE WORK
Olof Henricsson, Frank Bignone, Wolfram Willuhn, Frank Ade, Olaf Kübler
Emmanuel Baltsavias*, Scott Mason", Armin Grün"
Communication Technology Laboratory
Swiss Federal Institute of Technology (ETH)
CH-8092 Zurich, Switzerland
* Institute of Geodesy and Photogrammetry
Swiss Federal Institute of Technology (ETH)
CH-8093 Zurich, Switzerland
Commision lll, Working Group 2
KEY WORDS: Aerial Image Understanding, Feature Extraction, Building Reconstruction, DEM/DTM, Matching, Colour
ABSTRACT
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.
1 INTRODUCTION
The reconstruction of houses and other man-made objects
in 3-D is currently a very active research area and an is-
sue of high importance to many users of Geographic Infor-
mation Systems (GIS), including urban planners, architects,
and telecommunication and environmental engineers. Man-
ual 3-D processing of aerial images is time consuming and
requires the expertise of highly qualified personal and expen-
sive instruments. Therefore, the necessity to interpret, clas-
sify and measure aerial images and to integrate the results in
GIS is more urgent than ever. It is generally acknowledged
that good data is the most valuable and the most needed
component, prior to computer hardware, software and user
interface.
Methods for computer supported interpretation of aerial im-
agery have progressed in the wake of Computer Vision and
Digital Photogrammetry. The proceedings of the Ascona’95
workshop at Monte Verita, Switzerland gives a good account
of the current state-of-the-art [Griin et al. 1995]. The ob-
jective of the AMOBE project is to develop procedures for
extracting quantitative 3-D information of sparsely built-up
regions even under difficult terrain conditions. This goal pro-
vides complex technical and conceptual challenges and dis-
tinguishes the project from existing methods which work on
smooth terrain without being able to deal reliably with man-
made objects. The applications to Swiss scenery are immedi-
ate. Here, we present the strategies, current work, and some
ideas for future undertakings within the AMOBE project.
In section 2 we present our main strategies. Section 3 presents
the characteristics of the acquired data set. Section 4 deals
with digital surface/terrain models and color analysis to pro-
vide a means to detect and to provide a coarse description
of buildings. In section 5 we present our feature extraction
method and show how to relate pairs of contours to each
other by similarity in position, orientation, and their photo-
metric and chromatic region attributes. A novel approach to
reconstruct complex houses is presented in section 6; it in-
cludes segment stereo matching, coplanar grouping and mod-
eling in 3-D. Finally, we present some ideas of future work.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
2 STRATEGIES AND GENERAL FRAMEWORK
Although the research topics in the AMOBE project span a
large spectrum from Computer Vision to Photogrammetry,
attention is focussed on 3-D reconstruction of buildings, and
in particular on residential houses. Buildings are the most
predominantly and frequently occurring 3-D man-made ob-
jects in high resolution aerial images, and their reconstruc-
tion requires many components, such as camera models, im-
age processing, matching, texture and color modeling, geo-
metric processing and reasoning, as well as object modeling.
The employed imagery is assumed to be digitized photogram-
metric color photography. With this aerial imagery primarily
building roofs, and not walls, can be reconstructed.
The main features of our strategy for 3-D house reconstruc-
tion are illustrated in Fig. 1. The most important feature of
our strategy is the mutual interaction of 2-D and 3-D pro-
cedures at all levels of processing. This interaction is impor-
tant since neither 2-D nor 3-D procedures alone are sufficient
to solve the problems. Three-dimensional information, such
as Digital Surface Models and 3-D edges, should therefore
be derived as soon as possible. Because 3-D information is
available, object modeling can be done in 3-D, right from the
beginning. Whenever 3-D features are incomplete or entirely
missing, 2-D information should be used to infer the missing
information.
=
DSM generation feature
matching
coarse 3-D
building
detection
hypotheses
generation in 3-D
3-D reconstruction
hypotheses
generation in 2-D
2-D feature
color analysis ;
extraction
Figure 1: Strategy employed in the AMOBE project.
Several roofs of the residential houses in Fig 5A are neither
flat nor rectilinear, not even in object space. To reconstruct
the roofs of such complex houses, we have developed a proce-
dure that relies on hierarchical hypothesis generation in both
321
