ogies for 
vey  - 
try and 
software 
| digital 
levices - 
1 Greece 
solution 
scanner 
rnational 
Heritage 
nmagine 
A new 
mission 
he solid 
nposium 
the “last 
mm. V — 
ic ISPRS 
03 July, 
Antalya, 
Ministry 
concerns 
S eret 
(Digital 
orks for 
erson in 
Prof. C. 
bul 2004 
  
  
RECONSTRUCTION AND REPRESENTATION 
IN VIRTUAL REALITY OF A 3-D MODEL 
Hao-Hsiung Huang 
Department of Land Economics, National Chengchi University 
64, Sec. 2, Zhi-nan Rd., Wenshan, Taipei, 11623, Taiwan, R.O.C. 
hhh@nccu.edu.tw 
Commission V, WG V/2 
KEY WORDS: Three-dimensional Model, Virtual Reality, Reconstruction, Texture Mapping, Close Range 
ABSTRACT: 
3-D building models will be broadly constructed in a digital 3-D city map in the future. The models can be reconstructed using 
either aerial images or existing 2-D digital topographic map accompanied with height data. However, for large scale mapping, a 
more reliable and real 3-D model will be reconstructed using close range images. In this research, several close range images for a 
yard have taken by a calibrated digital camera. These images are used to reconstruct the yard model, also used for texture mapping 
of the yard. The resultant 3-D model has been represented in Virtual Reality Modelling Language(VRML). The visual appearance 
of the building model was high attractive. 
1. INTRODUCTION 
3-D building models have been attractive for a long while, 
especially when they have been represented in Virtual Reality. 
Usually, the models are reconstructed using satellite or acrial 
images or a scanned topographic map using height data. How 
accurate the reconstructed model is? Can it be suited for large 
scale mapping using a convenient digital camera? The primary 
purpose of the study has been to reconstruct a 3-D model using 
images taken by a digital camera. Experiments, generation of a 
3-D model, generation of virtual reality, and testing with 
ground true are described respectively in the subsequent 
sections. 
2. EXPERIMENTS 
2.1 Image Acquisition 
A Kodak DC280 digital camera was used for image acquisition. 
It is a non-metric camera, therefore, has been calibrated using 
the software, PhotoModeler Pro 4.0. The elements of interior 
orientation, including calibrated focal length, principal point 
location, radial lens distortion, and decentering lens distortion 
have been determined. 
An outdoor yard was selected as study object. Fifty five images 
were taken, ten of them were chosen for model reconstruction. 
Four images are shown in Figures 1, 2, 3 and 4 respectively. 
Geometric configuration and convergent angles have been 
considered while the pictures were taking. 
2.2 Ground Control 
Ground control has been performed in a local vertical 
coordinate system using a total station. For verifying the 
accuracy of the resultant 3-D model, 3 1check points were also 
surveyed in the same coordinate system. These points were 
well distributed throughout the test area. Marks have been 
stuck on these points before taking pictures, so that their 
corresponding images could be recognized easily. 
3. GENERATION OF A 3-D MODEL AND VIRTUAL 
REALITY 
The test area contains several different types of objects, object 
points, edges, curves, cylinders, lines, and surfaces. All of 
these objects have been processed in three steps to generate a 3- 
D model: 
I. The locations of objects are marked on each picture. 
2. The points are referenced together. 
3. The entered information is then processed. 
A 3-D model generated by the PhotoModeler can be exported 
and saved in the following types of file: Autodesk DXF, 
Microsoft DirectX, 3D Studio, Wavefront OBJ, VRML, IGES, 
and Raw. Usually, some or all of the following data can be 
exported: 3D points, point IDs, edges, lines, faces surface draw, 
curves and cylinders. 
Cosmo Player software supports the VRML file. The software 
can also be used for browsing a Virtual Reality Model in web. 
Therefore, the 3-D model built is exported and saved in a 
VRML file. 
Although a Virtual Reality Model can be observed in any view 
angle, only four models displayed in browser are shown in 
Figures 5, 6, 7, and 8 respectively. For comparing, Figure 5 
and Figure 1 have similar view angle, Figure 6 and Figure 2 
have