In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
EXTRACTING BUILDING FOOTPRINTS FROM 3D POINT CLOUDS USING
TERRESTRIAL LASER SCANNING AT STREET LEVEL
Karim Hammoudi, Fadi Dornaika and Nicolas Paparoditis
Université Paris-Est, Institut Géographique National, Laboratoire MATIS
73 Avenue de Paris, 94160 Saint-Mandé Cedex, France
{firstname.lastname} @ign.fr
KEY WORDS: 3D Point Cloud, Hough Transform, RANSAC Method, I\-means Clustering, Laser Scanner, Building Footprint,
Building Reconstruction, City Modeling.
ABSTRACT:
In this paper, we address the problem of generating building footprints using terrestrial laser scanning from a Mobile Mapping System
(MMS). The MMS constitutes a fast and adapted tool to extract precise data for 3D city modeling. Urban environments evolve over
time due to human activities and other factors. Buildings are constructed or destroyed and the urban areas are extended. Therefore, the
structures of the cities are constantly modified. Currently, building footprints can be generated using aerial data. However, aerial based
footprints lack precision due to the nature of the data and to the associated extraction methods. The use of MMS is proposed as an
alternative to perform this complex task. In this work, we propose an operational approach for automatic extraction of accurate building
footprints. We describe the challenges associated with the terrestrial laser raw data acquired in realistic and dense urban environments.
After a filtering stage on the 3D laser cloud point, we extract and reconstruct the dominant facade planes by combining the Hough
transform, the fc-means clustering algorithm and the RANSAC method. The building footprint is then estimated from these dominant
planes. Preliminary experimental results are presented and discussed. The assessments show that this approach is very promising for
the automation of building footprints extraction.
1 INTRODUCTION
Nowadays, city modeling has become an important subject of
research for architectural lasergrammetry, photogrammetry and
computer vision communities. There is an increasing need for
3D building descriptions in urban areas in several fields of ap
plication like city planning and virtual tourism. Therefore many
research activities on city modeling have focused on the auto
matic generation of 3D building models from aerial images. Most
pipelines which have been developed recover the 3D shape of
roof surfaces, but building ground footprints come from existing
databases acquired by the digitization and vectorization of cadas
tral maps or from surveying measurements.
Initially, the building footprints are extracted either in an auto
matic way using the aerial data (Cheng et al., 2008), (Tarsha Kurdi
et al., 2006) or in a manual way requiring many surveyors to make
measurements in the terrain. However, these footprint databases
sometimes do not exist (e.g., in less developed countries, etc.),
can be very difficult to obtain (e.g., in areas with difficult access
or prohibited overflights), or can even be of insufficient geomet
rical quality with respect to some applications. Moreover, the
automatic building footprints extraction using aerial images is a
hard task. Imprecise and/or incomplete focusing will affect the
modeling process in the sense that the final 3D building model
will lack accuracy and details.
Recent progress in technologies have allowed the development
and the construction of devices for rapid acquisition of 3D car
tographic terrestrial data with very high precision in urban envi
ronments. The Mobile Mapping System allows an easy coverage
of large scale areas such as districts and cities. The feasibility of
this kind of system has been demonstrated (Haala et al., 2008),
and the usage of this device is increasingly widespread for ap
plications like the conservation of patrimony (Baz et al., 2008)
or visualization. Many works using terrestrial laser scanning are
particularly focused on segmenting and texturing the building fa
cades (Boulaassal et al., 2007), (Pu, 2008).
This ground-based modeling is thus unavoidable for some ap
plications such as facade texturing where images acquired by a
ground based system need to be registered relatively to the aerial
3D model to ensure a satisfactory mapping. Matching the street
level images with the 3D aerial model is an extremely complex
due to the generalization problems. The data acquired by ground-
based 3D data collection systems, can be used to extract and
model facades that can advantageously replace the ground foot
prints in the aerial reconstruction process, thus leading to a co
herent use of both aerial and terrestrial data.
This paper focuses on the first step of a global 3D facade recon
struction framework, i.e. the extraction of the facade footprints
and planes. The MMS constitutes an alternative and reliable tool
which can be useful to obtain building footprints with very high
accuracy and details. The aim of this study is to propose an oper
ational approach for automated building footprints extraction in
urban environments. The remainder of the paper is organized as
follows: Section 2 states the problems related to the raw laser data
and their processing. Section 3 presents the proposed approach
for extracting the building’s footprints and facade planes. Section
4 gives some promising experimental results.
2 OVERVIEW ON PROBLEMS RELATED TO THE
LASER RAW DATA AND THEIR PROCESSING
In this study, we use a mobile mapping system for acquiring geo-
referenced 3D laser point clouds. The Terrestrial Laser Scanning
system (TLS system) is a 2D profile scanner. The third dimension
is induced by the vehicle displacement. In addition to this, the
Mobile Mapping System is equipped with a Global Positioning
System (GPS), an Inertial Measurement Unit (IMU) and a Dis
tance Measuring Instrument (DMI), namely an odometer. This
