on
lationships
ases) were
ie corridor:
ling, were
ed to this
coordinate
ich system,
f a simple
coordinate
4 & 5) and
; (figures 6
ion in order
office. The
ases of the
with their
ate the pa-
inks the va-
te system.
ordinate
d in bold)
Figure 8. 3D surface and photorealistic models of the corridor
and the classrooms or offices
32. Example of an outdoor scene
Although the method of 3D modeling proposed in this study is
intended for indoor scenes modeling, it can also be applied to
exterior scenes if their geometry is described by means of plane
surfaces. To show this possibility, this method was used to
establish the 3D model the old city hall of Zurich (Switzerland)
from images proposed by CIPA in a dataset (Streilein, 1999).
32.1. Steps of 3D single image modeling of the old
Zurich city hall
Used images
Two images, taken from the dataset, were used to carry out the
geometric model of the building. The first one covers the north
and east facades whereas the second covers north and west
facades. With regards to the south facade, it can be modeled by
means of points measured on the east and west ones.
Both selected images satisfy a perspective projection. This
allows the calculation of vanishing points associated to the
various axes of the local coordinate system.
Local coordinate system
To configure the scene for a single image modeling, a local
coordinate system was set up and control points were defined
(figure 9). This system must not be a right hand system. Four
control points were chosen on the north fagade, considered as
the reference plane of the scene.
Figure 9. Local coordinate system and control points
The same coordinate system was used to extract 3D coordinates
of points measured on both images indicated in figure 9.
Modeling with our application of single image modeling
To extract the 3D geometry of the Zurich city hall of by using
our application of single image modeling, the following stages
were applied:
l. Definition of two single image projects. Each project
requires an image, a file of control points and a camera file
(optional);
2. Computation of the vanishing points associated to the local
coordinate system axes. This is done by measuring a series of
parallel lines for each axis in the image;
3. Calibration of the camera by means of vanishing points
resulted from the previous stage;
4. Computation of the homography of the north fagade. This
is done with the four control points given in the dataset (figure
9);
5. Computation of the scale factor associated to the reference
direction(Z) by using a reference distance measured along this
direction;
6. Determination of the 3D coordinates of the points
measured on the various façades;
7. Establishment of a relational database by means of the
interface integrated into the application of single image
modeling. This database contains the semantic and topological
data of the building;
8. By means of the interface intended for topology-based
modeling and integrated into the main shape of a CAD system,
the visualization of building 3D surface model was done
automatically. In this database the WALL as a semantic type
represents the various facades whereas the semantic type
CEILING represents the faces of the roof;
Figure 10. 3D surface and photorealistic models
of the city hall of Zurich
By means of the CAD system, textures were attached to the
various faces and' the 3D-photorealistic model was generated
(figure 10). The textures of the various faces were calculated by
means of 3D points measured on these faces.
4. CONCLUSION
An approach has been proposed to realize interior and exterior
3d modeling of buildings from single images.
This approach can be summarized in the following points:
1. The approximation of the interior part as being composed
of simple and reference unit components;
2. The representation of the information in hierarchical form
3. The simultaneous data structuring in a relational database.
The RDBMS was chosen as the data processing tool.
Considering the geometric characteristics of the buildings, a
single image photogrammetry configuration was adopted as a
source of the 3D geometric data.
Three solutions were proposed to compute the 3D geometry
from one image using our application for single image
modeling, partly based on former researches. The 3D
reconstruction and visualization of the imaged scene are based
on the topologic relationships recorded in a relational database.
To validate our method, the method has been applied on an
indoor scene and an outdoor one.
—135—