3D BUILDING RECONSTRUCTION USING COMPOSITES OF SURFACE PRIMITIVES: CONCEPT
Scott Mason
Department of Surveying & Geodetic Engineering
University of Cape Town, South Africa
mason@engfac.uct.ac.za
Commision lll, Working Group 2
KEY WORDS: human settlement, mapping, reconstruction, aerial imagery, geometric modeling, generic building representation
ABSTRACT
A modus operandi for semi-automatic building reconstruction from photogrammetric imagery is presented. The key component
is a generic object modeling scheme based on composites of primitive surfaces. This scheme is well-suited to implementation
in an operational context. A strategy for an implementation is developed.
KURZFASSUNG
Ein modus operandi für eine halbautomatische Gebauderekonstruktion mit Hilfe photogrammetrische Bildaufnahme wird vorges-
tellt. Die Schlüsselkomponente ist ein generisches Objektmodellierungsschema, welches auf der Zusammenstellung primtiver
Fláchen basiert. Diese Schema ist für eine Anwendung in einem operationellen Zusammenhang gut geeignet. Ein Konzept für
die Implementierung wurde bereits entwickelt.
1 INTRODUCTION
One of the important tasks in extracting information from aer-
ial photographs is the precise, 3D reconstruction of buildings.
Applications requiring this data abound, from town planning,
architectural design and in-situ evaluation, real estate, virtual
tourism, dispatching systems for police and fire departments,
mobile phone communications, to environmental simulations.
Photogrammetric techniques for building reconstruction are
well-established but largely based on manual measurements.
While offering high accuracy and reliability, they require ex-
pensive instruments, highly-trained operators and are very
time-consuming. Since the early 1980's, research has aimed
at automating image measurement and understanding. Des-
pite considerable efforts, robust methods have failed to even-
tuate and formidable conceptual and technical problems re-
main. The key to improvements in operational performance
lie, therefore, in the development of semi-automation con-
cepts.
A fundamental issue in the design of semi-automatic proced-
ures is facilitating optimal utilization of the human's skills and
knowledge in an intuitive and economical manner. In building
reconstruction the user's primary task is to communicate the
presence and form (appearance) of the buildings in a given
scene. This can be achieved by selecting the model(s) best
describing the building and choosing a representation for that
model both amenable to automated feature extraction proced-
ures and natural to the user. We maintain that in attempting
to reconstruct a scene, especially for complex and irregular
structures, a user most naturally perceives a building as a
composite of surfaces. Since a photogrammetric image is a
projection of the 3D world into 2D, the visible surfaces are the
most direct source of information for reconstruction available
to the computer.
This paper presents, at a methodological level, a modus op-
erandi for semi-automatic 3D building reconstruction from
photogrammetric imagery. An object modeling scheme is de-
veloped which facilites a simple and intuitive user-interface,
enabling low-skilled users to communicate interpretations of
the shape of arbitrarily complex buildings. This scheme is
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
generic, modeling buildings as composites of primitive sur-
faces (CPS). In Sec. 2 basic issues in building reconstruction
from imagery are discussed. Sec. 3 reviews object models
for reconstruction and details introduces CPS modeling. Sec.
4 describes an operational building reconstruction strategy
based on CPS modeling. Finally, Sec. 5 outlines future work,
in particular, prospects for increased automation.
This research was commenced by the author under the
AMOBE project at ETH-Zürich and has been continued
within UrbanModeler, a project being conducted at UCT
to develop operational rapid mapping technologies for both
formal and informal settlements relevant to the African con-
text. Informal settlement mapping is of great relevance to
developing countries with drastic and rapidly changing urban
environments.
2 ISSUES IN BUILDING RECONSTRUCTION
2.1 Object Reconstruction Task
The terms object reconstruction, object recognition, and ob-
ject extraction are often used in the image understanding con-
text to denote the same or similar activities. We define re-
construction as the acquisition of a description of a specific
object. This is achieved by recognition, the task of finding
(object detection) and labelling parts of images(s) of a scene
that correspond to objects in the scene (Suetens et al, 1992).
Localization of the labelled parts is implied. In order to carry
out the recognition task, general descriptions, or models, of
each object to be recognized must be provided. Information
extracted from the image(s) by segmentation is matched to
these object models to enable labelling of the symbolic struc-
tures in the image(s). This information, which may include
object shape, pose, material, texture, colour and other attrib-
utes, constitutes a reconstruction of the object in the scene.
The object reconstruction process is illustrated in Fig. 1.
2.2 Data Sources
Currently, scanned, photogrammetric quality aerial photo-
graphy is the most practical source of imagery for accurate
and detailed building reconstruction. Image scale, film type
517
