DIGITAL
EXPERIENCE
S TEREOPHOTOGRAMMETRY
WITH
AN
EXPERIMENTAL
SYSTEM
Gerhard Koenig, Wolfgang Nickel, Juergen Storl
Technical University of Berlin
Photogramrnetry and Cartography
StraBe des 17. Juni 135
D-1000 Berlin 12, FRG
Commission II
ABSTRACT
An experimental Dig al Stereophotogrammetric System (DSS),
which is based on a commercially available image processing
system, has been set up at the Technical University of Berlin.
Any kind of stereoscopic imagery, such as scanning electron
microscope data, close range imagery, aerial photographs and
satellite image data, can be processed stereophotogrammmetrically.
This paper reports on the practical experience concerning
photogrammetric processing, especially of aerial photographs.
Due to hardware restrictions, so far only subsections of image
frames can be handled in reasonable time. The performance of the
system as well as the accuracy that could be achieved are
discussed and compared to the results obtained with an Analytical Plotter.
Based on the practical experience with the experimental DSS
conclusions are drawn concerning the design and architecture of
an operational system. An outlook is given to further system
development, considering new hardware components.
INTRODUCTION
At the ISPRS meeting of Commission II, WGII/2 in Baltimore 1986,
the authors reported on an experimental Digital Stereophotogrammetric System (DSS) which has been set up at the Technical
University of Berlin. This system then was suitable for
automatic processing of imagery reduced in size (512x512
pixels), such as digital scanning electron microscope images.
Based on the existing hardware described in (KOENIG, NICKEL,
STORL, 1986) the authors have implemented algorithms for the
photogrammetric processing of large data sets, especially of
digitized aerial photographs.
DATA ACQUISITION
For converting the analogue aerial photographs into raster data,
a color scanner of HELL company, Chromagraph DC 360, is used.
u
This system can handle transparency
s up to the format of
65x51 cm 2
a size which is limited by the diameter of the
rotating scanning drum. The aerial photographs can be digitized
in different resolutions. According to the experience of the
authors, a resolution of 25 ~ or 400 lines/em is sufficient for
stereophotogrammetric purposes. The advantage of higher resolution is too small to justi
its costs. This means that a
digitized stereo pair takes about 18 0 MB of disk storage .
Analysis of the scanned imagery has indicated that the Chromagraph scans with high quality and very little geometric distortions. The radiometric reproduction qual
is good so that
only few preprocessing steps are necessary for image improvement
that the quality of the photographs is sufficient.
ORIENTATION
The or
ation is done in two steps, i.e. relative orientation
and absolute orientation. After relat
ation the images
are resampled to the normal case of stereophotogrammetry. This
leads to a more rational automatic evaluation and enables stereo
vision of the whole model.
To overlook the scene both images are miniaturized to the size
of one screen. A 23 em x 23 em aer
photograph scanned with a
xel spac
of 25 microns leads to about 9000x9000 pixels.
20x20 p
ls are
and thus the whole scene can be
sualized at one t
The sampled miniaturizations are held
permanent
in two of the four image memories. The scene parts
to be actual
processed are defined
positioning a 25x25
l window and are loaded
the two rema
memories.
For each
(at least two of) the fiducial mark regions are
way. Fiducial mark matching then is performed by a
s (via an artificial template or via binarization and
calculation) . Image inner orientation is known and leads
to the
ers of an affine transformation which computes
image coordinates from pixel (i.e. scanner) coordinates.
RELATIVE
ENTATION
From s
up to any given number homologous points can be matched
interact
ly for relative orientation. This can be done either
defining templates in the first and search windows in the
second image or by using signalized object control points, if
there are any.
can be searched in one image and matched to
the other one or be searched independently in both images. Image
coordinates again are given by the aff
transformation mentioned above.
After this calculation resampling is carried out for the whole
stereo area (normal
us
bil
interpolat
) to eliminate
y-parallaxes. Thus the images are at hand for stereoscopic
vision by anag
c means and processing along horizontal
lar l
s.
ABSOLUTE
ORIENTAT
If the ground control points have not been measured for relat
ation this has to be done now. Us
the re
(i.e.
the relatively o ientated
s) instead of the
inal
images, stereos
c view of the earth surface is
the time this is realized in an anag
ic way. A three
dimens
1 float
mark is generated in the
memory and can be moved by a mouse in x- and ythe screen. Z direct
motion of the float
formed
press
two funct
buttons of the
the DSS can be handled like a convent
Control po
i.
model
have
which leads to an error
less than 0.5
accurracy
on the
tization rate of the
1
s.
a suff
accuracy of absolute orientation is
measur
least six ground control po
s. Us
s
absolute
orientat
resulted in standard plan deviat
s of 0. 7 m for
both the
ic Plotter KERN DSR 11 and the DSS. The standard
he
s were 0.04 m and 0.06 m, respect
scale ~ 1: 5000) .
3-D
PROCESSING
Automated evaluat
of a D
is a main
feature of the DSS. At present
is performed
f
homologue po
s in two
in an in ially preregular grid in object space.
algorithm uses the
Vertical L
Locus (VLL
method
BETHEL 1986)
to calculate
approximate hei
s. The f
correlation is done
a least
squares
ithm
1983)
In order to save
time there is no subpixel interpolation calculated
search and the center p
1 of the template
is set to the nearest ne
Thu
l result
does not represent a
object space
Normally
h ae al
failures can not be
been developed
These
ria
-
t
20%
orrelati
some cr e
have
in correlat
Desp e exist
lation may
ry
laxes caused
the standard
defined 1
of a parallax is
set for the actual
a pre-
the correlation coeffic
is
normal
0 8)
such results are
autocorrelat
funct
in the
a threshold value
calculat
the
of the
large
a cent
lax
fferences occur between
process
ous
wrong
research
failures and
the texture
is used as
a.
corre-
above result in
The detected
on the
on gross error and to correct
ly. Furthermore, he
s allowed
amount and the
of
calculation. A
or
the
DTM
on
different
s
cate that the strat
es to
correlat
results are very
and
40% o
excluded - even reliable correlation results. Accoris neccess
to de
more dens than the
one to have a su
c
number of po
for
lation If
s number is too smal
a
be calculated
s DTM
11 be used to evaluate
hei
s for re-start
the automatic process to
r
amount of homologue
Orientation
t
t
and correlation
needs about one second
and
po
evaluations o
te
1987) were 0.02 % of the
FUTURE
s
about the same
h
Anal
time
1 section
ze
DEVE
the
st
software tools
authors have
s
performance of ste
fferent sources. The ne
ies will
des
of an operational s
em consist
new hardware
s. This
em will be des
around a powerful work-
station which for performance purposes should be based on a
Reduced
struction
~ornputer
SC) that has a relat
ly
simple architecture In contrast to conventional
, a RISC
consists of relative
few
ructions name
se
ch are
used most
ly. Most of these instructions could be
a s
e and have a
xed format for s
have a large
ster set and are
compilers which leads to a
performance
rate around 10 MIPS
In
ion to such a
nents, as des
in
up a real t irne phot
process
architecture
performance needed
state of the
follows:
is
spe al hardware
, are necessary for sett
ric s
ern
The use of parallel
necessary in orde
to obtain the
ric applications
s
could be classified as
Instruction Mult le Data SIMD) processors, where each
at the same time, but
processor executes the same instruct
on
fferent data elements;
(MIMD) processors
which
- Multiple Instruct
Mult
where
each
processor
has
can be segmented in Mult
ers,
executes
s
its own attached memory and each
MIMD
processors
own program code, and Mult
share a common memory
a
ion
- and S
olic
where each processor
parallel
of an algor
ation in a p
s and
fa
. All
are s
zed to
at the same t
Modern image comput
systems make use of these arch ectures,
for example PIXAR (SIMD), or AT&T~s Pixel
(MIMD), which
is build around up to 82 Digital Signal Processors, rated at
lOMFlops each. These systems offer a
computational power
which is necessary for
and image processing applicat
s.
are equ
a
memory - up to
48MB are common - and in most cases could be configured
parallel transfer disks, which accomodate the storage of
data sets, such as aerial
, and
the
transfer rates needed for fast
processing
language 1
C-callable
for un
the task to
these
An advanced ste
ric workstation with h
formance should
well a colour mon
s
s
1 stereo di
use of 1
1 shutter
techno-logy. The use of a s
le specta le with polarizing
asses makes a comfortable and flicker-free view of the
stereoscopic images possible and seems to be the optimal
solution.
CONCLUSIONS
Experience has shown that processing of stereoscopic imagery by
use of conventional system architecture and photogrammetric
standard algorithms leads to satisfactory results concerning the
accuracy. Compared with conventional plotting in an Analytical
Plotter the algorithms implemented in the DSS for automated
evaluation are too time-consuming. At present a new photogrammetric workstation based on state-of-the-art technology is
under design, which will solve these problems.
REFERENCES
F., 1983: High Precision Digital Image Correlation.
Proceedings of the 39th Photogrammetric Week, Stuttgart, pp.
231-243.
ACKERMANN
J., 1986: Digitale Bildverarbeitung in der Nahphotogrammetrie - Neue Moglichkeiten und Aufgaben. Bildmessung und
Luftbildwesen, Vol. 54, pp.34-45.
ALBERTZ
J. ,
19 8 6:
The DSRll Image
Convention, Washington D.C., 6 pp.
BETHEL
Correlator.
ASP/ ACSM
A., 1987: Towards Real-Time-Photogrammetry. Invited Paper,
41th Photogrammetric Week, Stuttgart, 33 pp.
GRUEN
G., NICKEL W., STORL J., 1986: Processing of Scann
Electron Microscope Imagery in a Digital Stereophotogrammet c
System (DSS) . International Archives of Photogrammetry and
Remote Sensing, Vol. 26-2, pp. 130-138.
KOENIG
G., 1987:
Volle
geometrische
Systemkalibrierung
metrischer Luftbildkammern
Das Testfeld Brecherspitze.
Bildmessung und Luftbildwesen, Vol. 55, pp.151-154.
KUPFER