can be minimised if the inner orientation to the
analogue imagery is done using digital images. In
this case, the only significant instabilities are those
that occur while the analogue image is being
digitised. This was basically the philosophy
adopted while digitising for the investigation to be
described. The average observed drift of the
calibration translations during the digitisation of
one photograph was observed to be 0.5 |im. Two
values out of a total of thirty-two were observed at
over the 1 pm level, the maximum being 1.4 pm.
These figures relate to an average time period of
thirty minutes.
particularly experienced operator - in two
afternoons sessions. The coordinates were not
corrected for any distortions since these could be
added later if required.
4.4 Digital Measurement of the Block
4.4.1 Outline of the Procedure
The digital measurements were performed as an
interactive procedure, divided into two major parts.
In the first part, images of the signalised points are
acquired; the second part is the off-line matching of
these images to obtain the image coordinates.
4. Investigation into Aerial Triangulation
using Digital Point Positioning
4.1 The Aim of the Investigation
In this investigation, the S9AP with CCD cameras
will be used to relieve a human operator of the
point measurement task. Signalised points in a
photogrammetric test-block are positioned using
image matching and the calibration data as
described above. The investigation should shed
light onto two aspects: firstly how the accuracy of
digital and manual measurements compare, and
secondly where the potential problem areas would
be in a semi-automatic procedure.
4.2 The Test Area “Heinzenberg”
The data used for the investigation is taken from a
test-block flown in August 1986 in an area known
as Heinzenberg in eastern Switzerland. A sub
block of the original was used; the relevant details
are:
Area covered:
Ground height:
Flying height:
Camera:
Film:
Photo scale:
Overlap:
Block size:
6.0 x 6.0 km
650 m to 2150 m
4000 m
Wild RC10, 15/4 UAG
Kodak Panatomic-X
1 : 15,000 average
60% / 60%
4x4
The original film negatives were used; on average
there were 17 signalised points per image, with a
maximum of 25. The original block was measured
on the AC1 at the ETH in Zürich; the results of the
subsequent adjustments have been published (Grün
and Runge, 1987).
4.3 Manual Measurement of the Block
To get a meaningful comparison with manual
measurements, it is necessary to have datasets
which are in all other respects equivalent. Hence
the block was firstly observed on the S9AP using
the Phototriangulation Measurement software of
System-9. This was done by the author - not a
4.4.2 Digitisation of the Signalised Points
The image digitisation was done in a session-wise
fashion, whereby three separate sessions, usually
corresponding to a half-day, were required for the
16 photographs. Only the right-hand stage of the
S9AP was used. A session included frequent
calibrations of the CCD camera so that the results
relating to the stability could be obtained. The
instrument calibration was checked before and
after each session.
For a single aerial photograph, images of the four
fiducial mark images were first digitised, followed
by the images of the signalised points. The images
were stored on disk together with the stage
coordinates of the camera position. The time
required for digitisation of an image with, say, 20
signalised points was around half an hour, but this
was aided by a manual inner orientation and
knowledge of the image coordinates already. A
complete session of four photographs, including all
calibrations, was around three hours.
Image digitisation was done interactively with the
Maxvideo system and using software developed in
Zürich. For each image, the gain and offset levels
of the video signal digitisation were set manually,
so that, by visual inspection, the point of interest
(fiducial mark or signalised point) had optimum
contrast. This optimum depends on the density
levels in the aerial photography and hence also on
the S9AP illumination. In a few cases, it was also
necessary to change this. This all has important
implications for the potential of an automatic
procedure.
4.4.3 Point Positioning by Template Matching
Within the context of the collaborative project, the
constrained least squares image matching (LSM)
algorithm has been rewritten in the language C on
Sun Workstations. The algorithm is well known
and well documented in the literature (eg Grün and
BalLsavias, 1988); it is not the intention to elaborate
upon it here. For this investigation, the algorithm
was used in its unconstrained two-dimensional
mode, known as template matching as the one
39
