International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
The camera system EYSCAN used within our investigations
was developed by the KST GmbH in a cooperation with the
German Aerospace Centre. The system, which is depicted in
Figure 3, is based on CCD line mounted parallel to the rotation
axis of a turntable. Thus, the height of the panoramic image is
determined by the number of detector elements of the CCD line.
In contrast, the width of the image is related to the number of
single image lines, which are captured during the rotation of the
turntable while collecting the panorama. In our experiments,
this resulted in an image height of 10.200 pixels, while during a
360? turn of the camera more than 40.000 columns were cap-
tured. Since the CCD is a RGB triplet, true color images are
available after data collection. The spectral resolution of each
channel is 14 bit, the focal length of the camera is 60mm, and
the pixel size is 7um.
2.2 Geometric Processing
In order to map the visible faces of the buildings to the respec-
tive image patches of the panoramic scene, corresponding image
coordinates have to be provided for the 3D object points of the
building models. In accordance to the processing of standard
perspective images, the object coordinates X, are linked to the
corresponding camera coordinates x based on the well known
collinearity equation
Le.
RE Xk)
which defines a transformation between two Cartesian coordi-
nates system. In accordance to the approach described by
(Schneider & Maas 2003), a cylindrical coordinate system is
additionally introduced to simplify the transformation of pano-
ramic imagery. In this system, which is overlaid to the picture of
the EYESCAN camera in Figure 3, the parameter & represents
the scan angle of the camera with respect to the first column of
the scene. The radius of the image cylinder is given by the pa-
rameter r. In the ideal case, this parameter is equal to the prin-
cipal distance of the camera. The parameter n represents the
height of an image point above the xy-plane. Thus, this parame-
ter is related to the vertical distance of the object point to the
camera station. The transformation between the cylindrical
camera coordinates r,&,n and the Cartesian camera coordinates
is then given by
x=[xiy 2] =[r-cosé —r-siné j|
In the final step, the transformation between the cylindrical
camera coordinates and the pixel coordinates m,n is defined by
= thy
m=—
d,
Y
Similar to the processing of frame imagery, the pixel coordinate
n in vertical direction is determined by the corresponding com-
ponent n, ofthe principal point and the vertical resolution d, ,
which is defined by the pixel size. In contrast, the horizontal
resolution d, required to compute pixel coordinate m in hori-
zontal direction is defined by the rotation angle of the CCD line
per column during collection of the panoramic image.
The required exterior orientation parameters of the scene were
computed from a spatial resection. In order to allow for an in-
teractive measurement of the required control points the avail-
566
able 3D building models were approximately mapped to the
panoramic image. Based on this approximate mapping, a suffi-
cient number of visible faces, which were distributed over the
complete scene were selected and manually measured in the im-
age. In principle, the geometric quality of the EYSCAN camera
allows for accuracies of point measurement in the sub-pixel
level (Schneider & Maas 2003). Still, in our experiments only
accuracies of several pixel could be achieved. In our opinion,
this results from the fact, that the fit between model and image
after spatial resection is not only influenced by the geometric
quality of the image, but also from the accuracy, level of detail
and visual quality of the available 3D building models used for
the provision of the control points. While our urban model pro-
vides a reliable representation of the overall shape of the visible
buildings, the amount of detail is limited especially for the fa-
cades. As it is discussed earlier, this situation is typical for 3D
city model data sets, which are collected from airborne data.
Figure 4: Building models mapped to panoramic image.
Figure 4 exemplarily demonstrates the geometric quality of the
mapping process based on the result of spatial resection for a
part of the panoramic image.
Figure 5: Scenes generated using buildings with texture from
panoramic image.
Interr
Figur
image
textui
ate th
applic
objec
polyg
analy:
Usual
projec
specti
rectifi
these
proce
tions.
As it
ages (
from
lected
for vi
scene
will 1
minin
favou
essing
restriz
geoer
3.1.1
One i
ity of
While
ramic
meast
altern
collec
feasib
This t
al 20(
used i
buildi
expen
geore
a 3D
that c
be ref
the de
In orc
rect g
an ele
good
ber of
eas. S
config
Addit
multi;
GPS 1
10m
provi
