ay
bject
m in a
jn of a
" main
ues to
jn and
n bzw.
einen
Das
im die
m und
hoden
id die
. This
A first
orm to
y. We
of an
ing, a
esign,
pecific
in this
closed
g and
recise
made
range
ise 3D
tment,
taken
image
xd 3D
this
ion is
/ester-
a city
istoric
ormed
manually and based on traditional perspective
camera photography.
Till now no further works on the topic of indoor room
recording and reconstruction were found in the
literature study.
3 PRINCIPLES OF ELECTRONIC IMAGING
SYSTEMS
Digital photographs can be conventionally acquired
with film-based cameras and subsequent
digitization. Or electro-optical cameras combined
with solid-state sensors (see figure 1).
Close range photogrammetry seems to increasingly
employ electro-optical tools [Wester-Ebbinghaus,
1984], [Grün et al., 1993, 1995]. A variety of solid-
state imaging systems, from inexpensive
camcorders to specialized systems with high-
resolution sensors usually at a maximum
2000x2000 pixels are commercially available
[Luhmann, 1992]. One technique to increase the
resolution of such a commercial sensor is to move it
in the image plane. Only static objects can be
recorded, since the image is being acquired
sequentially from several partial images. Essentially
there exist micro-scanning cameras which offer a
resolution from 2300x3000 pixels [Lenz et al., 1990]
to 4500x3000 pixels [Richter et al, 1992]. With
macro-scanning the sensors are moved in multiples
of the sensor size, which results in a larger image
format. Merging of the partial images into a
seamless image arranges by precise mechanical
equipment [Poitz, 1992] or with a réseau
[Riechmann, 1990], [Leber et al., 1987].
An alternative to the use of square array sensors is
the use of linear CCD array sensors, that is being
transported across the focal plane of an optical
system. This is implemented in numerous
commercial products.
Film-based Laboratory
Camera Processing Digitalization
Eck 1
/ gum
/ LJ A
(0000
‚moon Anm
ZN pe a
Object | Digital 5.
ge —_— | Image |
Solid-state 3
Sensor
Figure 1: Two different methodes for generating digital
images
4 CONCEPT AND IMPLEMENTATION OF A
NEW „Indoor Room Imaging System"
Our attempt to reach a progress in the building of
3D models of a scene concentrates on two
objectives:
e reducing the manual work in building 3D models
55
e increasing the level of detail in such 3D models
Therefore we need a new and intelligent recording
strategie. A recording platform that acquires
electronic images of a scene by scanning across it
is necessary. From these image data we want to
extract the geometry and the photo-realistic texture
of the scene.
Area sensors which offer high geometric resolution,
e.g. 1000x1000 pixels, are very expensive. Today
tri-linear CCD arrays up to 8000 pixels each line are
commercially available at a fair price. In order to
design an imaging system, which offers a high
geometric resolution without cost explosion, we
decided to use a linear array sensor, a specific
camera and a camera motion equipment.
4.1 Basic principle
We propose an electronic imaging system, that is
based on a rotating camera with a linear array
sensor. It is a macro-scanning camera, where
merging of partial images into a seamless image is
based on a precise camera movement. CCD lines
are normally used in a linear motion to automatically
acquire electronic images at a high resolution in
close range photogrammetry. CCD based line
scanning technology is already used in remote
sensing [Ebner et al., 1992] [Hofmann, 1986, 1988]
[Wang et al., 1994]. We differ from these ideas by
using a rotational motion.
4.2 Image data representation
In our case the image data are acquired by sensor
rotation. Thus we obtain a cylindrical projection.
One advantage of a cylinder is that it can be easily
unrolled into a simple planar map. The surface is
without boundaries in the azimuth direction. One
shortcoming of a projection on a finite cylindrical
surface are the boundary conditions introduced at
the top and bottom.
4.3 Acquiring cylindrical projections
A significant advantage of a cylindrical projection is
the simplicity of acquisition. The only acquisition
equipment required is a line camera and a tripod
capable of continuous panning. Ideally, the
camera's panning motion would be around the
exact optical center of the camera lens system. In
practice, in a scene where all objects are relatively
far from the tripod's rotational center, a slight
misalignment offset of 20 millimeter causes an error
of about 0.025 pixel. Following from this example
we see that it is not necessary to rotate the line
camera exact around the optical center.
For convenience we define the sensors local
coordinate system (X’,Y’,Z’) in the center of
projection. The rotation of a CCD line around the
center of projection can be summerized in the
following mathematical model. À line camera with a
focal length f rotates around the Z'-axis by the angle
Qt. A point Pz(x,y,z) in the object space (figure 2) is
then mapped to the position z,,4 in the CCD line
according to the equations:
a — atn(y'/x') (1)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
