nbul 2004 
  
OCCLUSION-FREE 3D REALISTIC MODELLING OF BUILDINGS IN URBAN AREAS 
M. Varshosaz 
Department of Photogrammetry and Remote Sensing, 
Faculty of Suryveying, KNT University of Technology 
Vali-Asr Street, Mirdamad Cross, Tehran, Iran 
varshosazm@kntu.ac.ir 
KEY WORDS: Photogrammetry, Textures, Buildings, Modelling, Virtual Reality 
ABSTRACT: 
Currently the creation of textures for construction of Visually Realistic Models (VRMs) of buildings is a time consuming manual 
process involving the use of a number of separate software applications and substantial skill by the user. This paper presents the 
development of a novel procedure for the semi-automatic construction of textures, which makes use of a Charge Coupled Device 
(CCD) camera mounted on a motor-driven theodolite to acquire a digital texture database. The images are indexed using the angular 
horizontal and vertical readings from the theodolite. Given the coordinates of a building facade, an Automatic Texture Processing 
Tool (ATPT) finds the corresponding images covering the face and forms its texture. The system suggested here is substantially 
faster and easier to use compared the current methods available in the market. In addition, it can be used to easily merge textures of 
the same face in order to remove undesirable occlusions from textures. 
1. INTRODUCTION 
Developing Visually Realistic Model (VRM) of a building 
requires a 3D geometric model rendered with textures from 
building sides. This encompasses two distinct tasks namely the 
acquisition of building geometry and the processing of textures. 
Recent years have seen a number of developments in geometric 
modelling of buildings in 3D. The semi-automatic system 
developed at Bonn University (Gülch e/ al, 1998) is an 
example which aims on fast production of simple and 
complicated models of buildings in urban areas. 
The second, and difficult, task is the processing of textures. 
Textures are real images, taken from building façades, used to 
improve the visual impact of the models. They can also be used 
as a substitute to see/access the details that are missing in the 
geometric model. Processing textures for the photo-realistic 
documentation of buildings is carried out in two main steps: 
image acquisition and image processing. 
The aim of image acquisition is to provide digital input for the 
production of textures. Images can be obtained from either 
terrestrial or aerial photographs. Aerial photographs are used to 
cover details of roofs, while terrestrial images are used for the 
modelling of building façades. In rare situations, where the 
façades of buildings are not occluded, aerial photographs may 
also be used to extract textures of facades. However, due to the 
vertical nature of aerial photographs building facades are either 
invisible or subject to large perspective distortions. Therefore, 
the photo-realism of building facades can only be achieved if 
terrestrial images are acquired and mapped to their 
corresponding building faces (Brenner and Haala, 1998). 
Once the images are captured they have to be processed in 
order to form textures. This includes setting up the 
correspondence between the geometry and texture, rectifying 
and mosaicing the images, and enhancing the resulting textures 
by adapting the colour, intensity, saturation, and clearing 
occlusions shadows as required. To create textures, the first step 
437 
is to set up the correspondence between the image and object 
coordinate systems. As the interior and exterior orientation of 
images is generally unknown, usually this is done manually by 
measuring the image coordinate of few control points whose 
positions on the ground are known. Depending on the type of 
the camera used, the images may show a panoramic or 
perspective view of objects, which need to be compensated for 
through a rectification process. The images covering a 
particular building face are then mosaiced to form the complete 
texture of the face. 
Occlusion by static or moving objects is another problem in 
texture processing. The effect of occlusions can be reduced by 
having a good plan prior to data acquisition to gain an 
obstruction-free view of the buildings. A partial solution to 
remove occlusions is to apply computer graphic methods filling 
in the missing parts with artificial or image-based templates. 
Techniques like this, however, may result in unrealistic 
impression of objects in the model. Besides, even with imagery 
of similar information to that missing, different intensity, hue, 
etc. may result in discontinuities over the created texture. 
Looking at the literature, it can be realised that there exist no 
fully automatic texture processing technique. Attempts so far 
have mainly concentrated only on the automatic acquisition of 
images. The modelling process is highly dependent on human 
interaction (Debevec, 1999) and, thus, improvement in this area 
is extremely desirable (Maresch, 1997). 
The system presented here is an attempt to simplify and speed 
up the texture-processing task. It uses a CCD camera to provide 
digital input for textures, which are then processed 
automatically. The details of how the system works can be 
found in Varshosaz (1997). The aim here is to briefly review 
the main components of the system and show how it can be 
used to remove occlusions from textures.