rery promis-
rmation like
e increasing
opinion the
a bases like
1e a topic of
W., Stein-
]ümmer, L.
lding recon-
109-118.
‘Extracting
tern Analy-
13.
hetic land-
d visualiza-
togrammet-
e, pp. 293-
yuter vision:
EEE Trans-
Intelligence
using linear
>, eds, ‘Pho-
ann Verlag,
for the de-
n À. Gruen,
> Extraction
ace Images’,
n, pp. 211=
on Karten-
en aus der
on, Univer-
SEMI-AUTOMATIC MEASUREMENT OF SIGNALIZED GROUND CONTROL POINTS AT DIGITAL
PHOTOGRAMMETRIC WORKSTATIONS
Michael Hahn, Michael Kiefner, Antje Quednau and Esther Hinz
Institute of Photogrammetry
Stuttgart University
PO.B. 106037
70049 Stuttgart / Germany
Intercommission Working Group lI/IlI
KEY WORDS: Control point library, measurement of signalized points, semi-automatic process, photogrammetric worksta-
tion, data compression
ABSTRACT
The measurement of ground control points is the weakest point in the automation of photogrammetric orientation pro-
cedures. Automatic interior and relative orientation procedures as well as point transfer modules for aerial triangulation
are now available in Digital Photogrammetric Systems. What is still missing are automatic exterior orientation processes
because of the lack of a reliable procedure for ground control point measurement.
The paper presents an investigation into the semi-automatic measurement of signalized points. We propose a concept
which is based a library of templates with various types of signals. The interactive part of the process is the coarse mea-
surement of a signal by a human operator. With this approximate location the precise measurement is solved by matching
taking different templates into account. If a signal is imaged in several photographs a multi-image solution is derived even
though the starting value has to be given only in one image. S
pecial attention is drawn to the varying background problem.
An experimental investigation is carried out with 36 photographs in which 1714 image points of the imaged signals have
to be measured. The experimentally found rate of over 99% of successfully localized signals has exceeded all prior ex-
pectations. Furthermore, the accuracy of this automatic measurements of 2.8 um is 15% better than the 3.2 um accuracy
obtained for the manual measurements. Further, dependency on the image compression rate and on the type and shape
of the templates is reported.
1. INTRODUCTION
The automation of photogrammetric image orientation pro-
cedures has now been a central topic of research for many
years. Today, algorithms for solving the interior and the rela-
tive orientation are quite successful and are now integrated
in Digital Photogrammetric Systems. Similar progress can
be observed in the development of point transfer proce-
dures for automated aerial triangulation. The success of
these procedures is substantially a success of image anal-
ysis and image matching. In particular, the extraction of
prominent points in images and feature based or area
based matching of those points are used for the develop-
ment of robust and reliable processes. In interior orientation
the precise measurement depends on matching the image
with a more or less binary pattern of the fiducial mark.
The development of algorithms for reliable location of
ground control points has been slow over the years. In
present work the user of a Digital Photogrammetric Sys-
tem has to measure these points in some manual mode.
The general problem in this case is to establish the corre-
spondence between the points on the ground and the im-
ages of those points. For automatic processes this 3D to 2D
correspondence problem is significantly more difficult than
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
the 2D to 2D correspondence problems mentioned above.
One reason is that classical coarse-to-fine processing us-
ing image pyramids can not be applied and another suc-
cessful method for restricting search space and producing
approximate values for solving the correspondence prob-
lem is not available. A second reason is that in particular
natural ground control points, e.g. street crossings, in gen-
eral appear in a variety of different shapes. Referring to this,
the classical discs used for signalization of ground control
points are quite simple. For this uniformly shaped targets
the chance to be successful in detection and location of
the signals increases. But corruption of signals, inhomo-
geneous background and the usually small size of those
signals are problems which make automatic measurement
difficult.
A general formulation of a photogrammetric process for au-
tomatic ground control point measurement has to take care
of the scene interpretation problem. Often sketches are
prepared and comments are added to assist an operator in
finding the ground control points in the photographs. With-
out such help it would be also a big problem for a human
operator to scan through an image and detect signalized
points. The transfer of an operator's knowledge and in-
terpretation capability to a procedure which applies rules
