Elements of the corridor are walls, windows and doors. This 
geometry is the ideal environment to apply our modeling 
method. However, the weak distance between the object and the 
camera, has made the data extraction and recording rather 
difficult. 
Corridor dimensions are approximately of (2.2m x 50.0m). 
  
Figure. 4. Test building (ENSAIS) 
3.1.1. Modeling data 
Shots were made with a non-calibrated digital camera (FUJI DS 
300) and seventeen images were necessary to cover the corridor. 
Some control distances were measured in the reference plane (in 
our example the plane XY of a local coordinate system). 
Distances along the (Z) axis were also required to determine the 
scale factor associated to this axis. 
Additional measurements, realized by means of a simple ribbon 
and directly made on some parts of the corridor, were necessary 
to complete the 3D model. 
3.1.2 Modeling steps 
Defining a local coordinate system 
A local coordinate system was defined in object space. The 
reference plane (XY) is defined as the corridor’s ground plane. 
The axis (Z) was chosen as a perpendicular direction to the 
reference plane (XY). 
Extraction of the 3D geometric data 
Extraction of the 3D geometric data was made by means of our 
application of single image modeling. First of all, vanishing 
points associated to the various axis of the local coordinate 
system were computed. Then, the homography parameters of 
the reference plane were determined using control points. 
Extracted 3D geometric data are recorded in a text file. This file 
constitutes the node table in the resulted partial database. 
Extraction of the topologic and semantic data 
During the modeling, the operator extracts simultaneously the 
geometric, topologic and semantic data of a given surface. He 
specifies the numbers of points, which constitute the limits of 
this surface as well as its semantic type. These data are recorded 
in a database generated automatically. This database is then 
used with a GUI integrated into a CAD system to reconstruct 
the 3D model. 
Data Combination and 3D surface model generation 
The 3D geometric data associated to the topologic relationships 
and semantic types (recorded in four partial databases) were 
used to generate the following 3D surface model of the corridor: 
Adding classrooms and offices 
The same steps, applied during corridor modeling, were 
followed to model classrooms and offices attached to this 
corridor. For each classroom and office, a local coordinate 
system has been set up. In the reference plane of each system, 
some control points were measured by means of a simple 
ribbon. 
Classroom and offices were modeled in their local coordinate 
systems using single image photogrammetry (figures 4 & 5) and 
components were merged thanks to the virtual doors (figures 6 
& 7). 
  
Figure 5. Two shots only are used for 
the modeling of one room 
     
  
Figure 6. 3D model of a classroom 
(computed from two single images) 
The virtual doors were specified during data extraction in order 
to know which door belongs to which classroom or office. The 
tables, which represent theses doors in the databases of the 
different classrooms and offices, were then used with their 
corresponding in the corridor's database to calculate the pa- 
rameters of the 3D similarity transformation which links the va- 
rious local coordinate systems to corridor's coordinate system. 
     
ie 
i 
1 
' 
1 
! 
! 
1 
  
  
Figure 7. Linking a classroom to the corridor's coordinate 
system using the virtual door (indicated in bold) 
-134- 
Figi 
32. 
Altho 
inten 
exteri 
surfac 
estab 
from 
3.2.1. 
Used 
Two 
geom 
and € 
facad 
mean 
Both 
allow 
varioi 
Local 
To c 
coord 
(figur 
contr 
the re 
The s. 
of poi 
Mode 
To ex 
our aj 
were ; 
1. ] 
requir 
(optio