ll dee 
  
FUSION OF TERRESTRIAL LASER SCANNER DATA AND IMAGES 
FOR BUILDING RECONSTRUCTION 
Impyeong Lee *, Yunsoo Choi 
Dept. of Geoinformatics, University of Seoul, Seoul 130-743, Korea, (iplee, choiys)@uos.ac.kr 
KEY WORDS: Fusion, LIDAR, Image, Building, Reconstruction, Registration, Urban, Laser Scanning 
ABSTRACT: 
The increasing need and use of 3D GIS particularly in urban areas has produced growing attention on building reconstruction, which 
attempts to generate geometric and radiometric models of buildings usually from sensory data. The popular data used for building 
reconstruction are traditionally aerial or satellite images, more recently airborne LIDAR data, or the combination of these data. 
Extensive studies on building reconstruction from these data have developed some competitive algorithms with reasonable 
performance and some degree of automation. Nevertheless, the level of details and completeness of the reconstructed building 
models often cannot reach the high standards that is now or will be required by 3D GIS in future. Hence, the use of close-range data 
that can provide higher resolution and more complete coverage has been intensively emphasized. The representative close-range 
sensors are digital cameras and terrestrial laser scanners providing images and points, respectively. We attempted to develop a fusion 
approach for building reconstruction from both points and images. The proposed approach was then applied to reconstructing a 
building model from real data sets acquired from a large complex existing building. Based on the experimental results, we assured 
that the proposed approach cam achieve high resolution and accuracy in building reconstruction. The proposed approach can 
effectively contribute in developing an operational system producing large urban models for 3D GIS with reasonable resources. 
1. INTRODUCTION high precision and complete coverage. Hence, the use of close- 
range sensors has been recommended in such cases. 
The next generation applications such as Telematics, LBS 
(Location Based System), and Ubiquitous has been rapidly The most representative close-range sensors are digital cameras 
developed in recent years. As the most crucial basis for such and laser scanners nowadays. The cameras produce images 
applications, the importance of 3D GIS has been increasingly while the scanner points. The images provide better information 
emphasized. One of the important components for 3D GIS is about the edges and corners of buildings while the points about 
considered as urban models; and hence their acquisition and the facets. In addition, the reconstruction processes starting 
periodical update are inevitable to maintain operational 3D GIS. from images were not easily automated while those from points 
For reference, the current status and perspectives on 3D GIS are more possibly automated. More elaborate comparison results 
presented by Stoter and Zlatanova (2003) and Zlatanova et. al. are presented by Ackerman (1999), Baltsavias (1999), and 
(2002). Schenk (1999). Hence, to benefit from such complementary 
properties, it is more desirable to employ a fusion approach that 
Urban models mainly contain models of artificial structures use both points and images simultaneously for building 
such as buildings and roads. Many studies have then focused on reconstruction. 
extracting buildings and roads from sensory data in automatic 
or semi-automatic ways. Particular interests have been in Based on the careful review on the research trends and the 
building reconstruction, which attempts to generate building ^ needs of applications available now or to be flourished in future, 
models usually from sensory data. The building models usually we argue that operational systems should be developed to 
include the geometric descriptions (shapes and positions) and perform building reconstruction. from images and points 
radiometric descriptions (texture) of a building. The sensory acquired by close range sensors in interactive or nearly- 
data used for building reconstruction have been mostly aerial or automatic ways. Hence, the purpose of this research was to 
satellite images, more recently airborne LIDAR data, or the perform a pilot study toward the development of such systems. 
combination of these data (Halla and Brenner, 1999; Maas and We thus established a conceptual framework for such systems 
Vosselman, 1999; Jung, 2004; Suveg and Vosselman, 2004). and the core processes constituting the framework. We then 
validated this proposed framework and processes by applying 
The previous studies on building reconstruction have developed them to reconstructing a building model from real data sets 
some competitive algorithms with reasonable performance and acquired from a large complex existing building. 
some degree of automation. Nevertheless, the level of details 
and completeness of the buildings reconstructed using such This paper introduces the proposed framework consisting of 
algorithms often cannot reach the high standards that is now Or two main stages, data acquisition and data processing in section 
will be required by 3D GIS in future. This problem mainly 2. It then describes each stage in section 2 and 3, respectively. 
originates from the limitation of the sensors themselves. That is, Section 4 presents the experimental results from the application 
since most sensors acquire the data in far-range, for example, to real data sets. Finally, this paper concludes with summary 
from air or space, they cannot provide the data of sufficiently and some remarks in section 5. 
SES 
* Corresponding author. 
984 
Ini 
We 
fron 
fran 
proc 
data 
imag 
digit 
invo 
data. 
posi 
(text 
3.1 
3.1.1 
num 
build 
andr 
It th 
Conv 
The 1 
by th 
of thi 
Coorc 
laser 
and 
coord 
313 
acqui 
neces 
accur. 
since 
To de 
shoul 
algori