-B3, 2012
| edge detection
both used in our
atching result is
ition of covers
e detection and
method, such as
e.
ly supported by
C and Ho, PC,
he underground
Computing and
nal Conference
proved classical
’s detection and
2006), pp. 504-
andom Sample
Applications to
ACM, 24(6), pp.
f automatically
overs based on
h of Computers
detection with
Pattern Analysis
5.
nhole detection,
International
1, pp. 188-195.
ation for urban
552-1555.
ipherical.asp. (2
AUTOMATIC 3D BUILDING MODEL GENERATION FROM LIDAR AND IMAGE
DATA USING SEQUENTIAL MINIMUM BOUNDING RECTANGLE
E. Kwak', M. Al-Durgham °, A. Habib *
? Dept. of Geomatics Engineering, University of Calgary, 2500 University Drive, Calgary, T2N 1N4, AB, Canada -
(ekwak, ahabib)@ucalgary.ca
® Dept. of Civil Engineering, University of Toronto, 35 St. George Street, Toronto, M5S 1A4, ON, Canada -
mohannad.al.durgham@utoronto.ca
Commission III, WG III/4
KEY WORDS: LiDAR, Photogrammetry, Building, Modeling, Automation
ABSTRACT:
Digital Building Model is an important component in many applications such as city modelling, natural disaster planning, and
aftermath evaluation. The importance of accurate and up-to-date building models has been discussed by many researchers, and many
different approaches for efficient building model generation have been proposed. They can be categorised according to the data
source used, the data processing strategy, and the amount of human interaction. In terms of data source, due to the limitations of
using single source data, integration of multi-senor data is desired since it preserves the advantages of the involved datasets. Aerial
imagery and LiDAR data are among the commonly combined sources to obtain 3D building models with good vertical accuracy from
laser scanning and good planimetric accuracy from aerial images. The most used data processing strategies are data-driven and
model-driven ones. Theoretically one can model any shape of buildings using data-driven approaches but practically it leaves the
question of how to impose constraints and set the rules during the generation process. Due to the complexity of the implementation of
the data-driven approaches, model-based approaches draw the attention of the researchers. However, the major drawback of model-
based approaches is that the establishment of representative models involves a manual process that requires human intervention.
Therefore, the objective of this research work is to automatically generate building models using the Minimum Bounding Rectangle
algorithm and sequentially adjusting them to combine the advantages of image and LiDAR datasets.
1. INTRODUCTION
The importance of up-to-date and accurate geospatial
information has been emphasized with the increasing demand
for Geographic Information Systems (GIS). Digital Building
Model (DBM) is one of the important components among the
geospatial information especially in urban areas. They are
required as an input in many applications such as city
modelling, natural disaster planning, and aftermath
evaluation. With the development of sensor technology and
the increase of user requirements, many different approaches
for efficient building model generation have been proposed
(Rottensteiner et al., 2005; Habib et al., 2010; Huang and
Sester, 2011). They can be categorised according to the data
source used, the data processing strategy, and the amount of
human interaction (Vosselman and Maas, 2010).
In terms of data source, aerial imagery and LiDAR data are
among the most commonly used sources to obtain 3D
building models which can provide good vertical accuracy
from laser scanning and good planimetric accuracy from
aerial images. Due to the limitations of using single source
data, integration has been already recommended since it
preserves the advantages of the involved datasets. Using only
aerial imagery provides reliable results based on a
photogrammetric approach, but the low degree of automation
during the matching process is a main limitation especially in
case of occlusions. Partial or complete occlusions are
common problems for images over urban areas. LIDAR
systems provide direct 3D positional information eliminating
the need for a matching process, but the derived boundaries
do not represent the actual building boundaries due to the low
sampling rate of airborne LiDAR data. This makes LIDAR
* Corresponding author.
data not sufficient as a stand-alone source. The advantages of
the integration of LIDAR and image datasets for the building
reconstruction are already well summarised in many
researches (Cheng et al, 2008; Demir et al, 2009;
Awrangjeb et al., 2010; Habib et al., 2010).
The most used data processing strategies are the data-driven
and the model-driven ones. Data-driven approaches which
are also called bottom-up processes often rely on LiDAR data
and do not make assumptions regarding the building shapes.
Theoretically one can model any shape of buildings using
data-driven approaches, but practically it leaves the question
of how to impose constraints and set the rules during the
generation process (Brenner, 2005). Due to the complexity of
the implementation of the data-driven approaches, model-
based approaches, i.e., top-down processes, draw the
attention of the researchers. It predefines building models
using model parameters, and the model parameters are
updated using information derived from existing data.
Complex building models can be constructed by combining
small sets of model primitives depending on the desired level
of detail. While it provides robust computation, establishing
the representative models requires manual interaction (Tseng
and Wang, 2003). Therefore, the objective of this research
work is to automatically generate building boundaries using
the minimum number of models while combining the
advantages of image and LiDAR datasets. In this paper, the
focus of reconstruction is on complex structures, which
comprise a collection of rectangular primitives. The
assumption is that most of the existing buildings, especially
in urban areas, can be reconstructed using combination of
rectangular shapes. To meet this objective, the proposed
