687 
LARGE SCALE TEXTURE MAPPING OF BUILDING FACADES 
Y. K. A. Tan 3 ' *, L. K. Kwoh a , S. H. Ong b 
a Centre for Remote Imaging, Sensing and Processing, National University of Singapore, Block SOC1, Level 2 Lower 
Kent Ridge Road, Singapore 119260 
b Dept, of Electrical and Computer Engineering, National University of Singapore, Block E4, Level 5, Room 14 
Engineering Drive 3, Singapore 117576 - crstyka@nus.edu.sg 
KEY WORDS: Virtual Reality, Texture, Photo-realism, Building, Visualization, Photogrammetry. 
ABSTRACT: 
A virtual 3-dimension city model is a sophisticated application of geoinformatic systems as it represents layouts, activities and 
functionalities of a real-world community. To enable human users to better visualize their living spaces, a virtual reality city model 
must be both immersive and accurately re-created. Visualization of city models in such environments is a combination of many 
challenging and laborious tasks, one of which is the texturing of building models façade. To improve the photo-realism and textural 
information of the building model façade, close-range digital photographs of buildings are taken at oblique angles and used to 
texture the building models. This paper will show that by taking oblique views of buildings, we are able to perform large scale 
texture mapping on multiple buildings models and improve the overall workflow of the entire façade texture extraction process. 
Registration of the digital images must be carried out to obtain the images’ extrinsic and intrinsic parameters before accurate texture 
mapping can take place. Spatial distortions due to irregularly shaped façade polygons are introduced during the texture extraction 
process and a method was devised to overcome this problem. A test example will be used to demonstrate the system of performing 
large scale texture mapping using minimal amount of close-range photographs. 
1. INTRODUCTION 
A 3-D city in virtual reality is a sophisticated application of 
geoinformatic systems as it is a representation of layouts, 
activities and functionalities of a real-world community. It is an 
integrated effort in the fields of computer graphics, remote 
sensing and engineering to achieve a feasible, yet convincing 
representation of the city in the virtual world. An accurate and 
immersive virtual reality city model would enable the users to 
better visualize their living spaces interactively with greater 
detail and thus model human activity more intuitively and 
efficiently than before. All these would depend on the accurate 
reconstruction on both the 3-D terrain and building models as 
well as the correct texture mapping of model surfaces. 
Current generation satellite imagery coupled with Rational 
Polynomial Coefficient (RPC) can be used to accurately 
estimate 3-dimensional building information (Grodecki, 2001). 
In the case of IKONOS stereo images, the accuracies are within 
1 metre horizontally and 2 metres vertically hence acceptable 
for use in urban mapping whereby landscape and building 
objects are usually bigger than 3 metres (Grodecki and Dial, 
2001). With the advancements in building extraction techniques, 
we are able to extract a large number of buildings in a short 
period of time. For example, using a technique develop at 
CRISP (Huang and Kwoh, 2007), we are able to extract about 
60 buildings per hour in an urban area using a single IKONOS 
image coupled with RPC and building shadow information. 
Remote sensing data can provide limited texture information to 
the extracted building models as this is restricted by the sensor 
look angles. Very often, only the building model tops and 
certain building model facades are textured and the building 
model facades extracted from satellite imagery are pixelated 
due to the lack of resolution and texture information (Huang et 
al., 2006). This is because the represented skew area of building 
façade which can be seen from the satellite imagery is relatively 
small when compared to the actual size. 
3-D virtual reality city model visualization platforms use 
OpenGL or DirectX to effectively render the buildings models 
and ground terrain using hardware acceleration. When texture 
information of the building facade is not known, it is a common 
practise to assign a pseudo texture to the building model instead 
(Beck, 2003). Although this provides a better presentation of 
the building model, it is not ideal as it does not portray the true 
characteristics of the building façade. Such pseudo 
representations of the building facades, makes the overall 
visualization system less realistic and less useful. Hence, to 
improve the overall photo-realistic quality and usability, we 
texture the building façade with terrestrial digital photographs 
taken using commercially available cameras. However, the 
process of allocating and mapping the appropriate texture from 
terrestrial photographs to correct building façade is both 
laborious and time-consuming. To address this issue, we have 
developed a system which implements a semi-automated 
process for the large scale texture mapping of the building 
façade and this will be presented in this paper. 
2. TEXTURE MAPPING OF BUILDING FACADE 
The texture mapping process of building façade involves the 
transformation from image space to building model space. 
There are two methods of carrying out the transformation. 
Firstly, the forward mapping method selects the appropriate 
texture polygon from the image and maps it to the matching 
building façade. Conversely, the reverse mapping method 
* Corresponding author.