713 
BUILDING DETECTION AND RECONSTRUCTION FROM AERIAL IMAGES 
Dong-Min Woo a ' *, Quoc-Dat Nguyen 3 , Quang-Dung Nguyen Tran 3 , Dong-Chul Park 3 , Young-Kee Jung b 
a Dept, of Information Engineering, Myongji University, Gyeonggido, Korea - (dmwoo, datnguyen, qdungtran, 
parkd)@mju.ac.kr 
b Dept, of Computer Engineering, Honam University, -Kwangju, Korea - ykjung@honam.ac.kr 
Commission III, ThS-7 
KEY WORDS: Building, Detection, Reconstruction, Aerial, Image, Feature 
ABSTRACT: 
This paper presents a new method for building detection and reconstruction from aerial images. In our approach, we extract the 
useful building location information from the generated disparity map to segment the interested objects and consequently reduce 
unnecessary line segments extracted in low level feature extraction step. Hypothesis selection is carried out by using undirected 
graph, in which close cycles represent complete rooftops hypotheses. By using undirected graph, hypothesis selection becomes a 
simple graph search for close cycles. This significantly improves the performance of the system over the traditional hypothesis 
selection methods. We test the proposed method with the synthetic images generated from Avenches dataset of Ascona aerial images. 
The experiment result shows that our method can be efficiently used for the task of building detection and reconstruction from aerial 
images. 
1. INTRODUCTION 
The building detection and reconstruction from aerial images is 
one of the challenge tasks in computer vision. It has been used 
widely in various applications including traditional applications 
such as cartography and photo-interpretation and recent 
application including mission planning, urban planning, 
computer graphics and virtual reality. There are two main 
problems needed to solve in any buildings detection approach. 
The interested objects need to be segmented from the 
background, and the fragmented line segments of the interested 
objects should be grouped to human-made structures. These 
tasks are very challenging, because the objects of interest could 
be partly occluded by the presence of vegetation, shadows, road 
and other objects. Moreover, lines and comers of object are 
often fragmented and missed due to the typical failures of low 
level features extraction. 
Early approaches tried to use a singe image only (Huertas, 
1988; Lin, 1998). Since the inference of 3D information from 
one image is very difficult and there are still some ambiguities 
in the detected buildings that can be only resolved by feature 
matching in multiple images, the application of the single 
image approach is very limited. In this context, most of the 
recent work in this area has focused on the multiple-view 
analysis (Fischer, 1998; Noronha, 2001; Collins, 1998). Mohan 
and Nevatia (Mohan, 1989) proposed an approach for detecting 
and describing buildings in aerial image using perceptual 
grouping. They demonstrated the usefulness of the structural 
relationships called collated features which can be explored by 
perceptual organization in complex image analysis. All 
reasonable feature groupings are first detected and the 
candidates are then selected by a constraint satisfaction network. 
But this approach involves all extracted line segments in the 
image. Consequently it costs a big computational effort. It also 
depends on the accurate extraction of line segments. 
Some approaches such as Lin (Lin, 1998) and Noronha 
(Noronha, 2001) used hypothesis and verification paradigm 
based on perceptual grouping. Hypotheses are generated by a 
hierarchical perceptual grouping process and verified by the 
evidence of visible walls and expected shadows. But the system 
needs to make several decisions in the selection and verification 
process based on simplicity and intuitive judgments. In 
monocular analysis, Jaynes (Jaynes, 1994) proposed feature 
relation graph, in which hypothesis selection takes places as a 
graph search problem. This approach improved the performance 
of hypothesis selection step. However, it is limited on 
rectangular buildings and tends to generate false hypotheses in 
complexity images. 
In this paper, we extract the suspected building regions in the 
disparity map generated from aerial images and utilize them to 
get the location of interested objects in the image. The 
suspected building regions are areas, where pixel values rapidly 
changes relative to the surround area. This process can reduce 
the unnecessary line segments from the low level feature 
extraction result. Also, we employ perceptual grouping to build 
collated features, which are used to generate rooftop hypotheses. 
This perceptual grouping process removes the unsatisfied line 
segments during the grouping process. Comers and line 
segments in collated features are used to build the undirected 
feature graph, in which close cycles are detected as rooftops. 
With the open connectivity rules between nodes in the graph, 
our approach can effectively detect rectilinear shape buildings. 
The remainder of this paper is organized as follow: an overview 
of the system is shown in Section 2. Section 3 describes the 
generation of suspected building region. In Section 4, the 
principle of low level features processing and rooftop 
hypothesis is introduced. Section 5 presents experimental 
results on an aerial image data set. Conclusions are given in 
Section 6. 
* Corresponding author.