ınbul 2004
quasigeoid
1e (Denker
goes from
; against a
mal has a
and in the
jviation of
lues. This
ed ground
le 1: 5000
p to 4.4cm
yrth-south-
it can be
ent if the
final effect
few mm.
59°24'
59°
Vrorr
e OEEPE-
ing in the
etween the
t areas is
area to the
km if the
ve shown
over 30km
blems are
AU to the
iation are
:quired for
before and
blished by
he Institut
University
c Institute
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
together with the National Land Survey of Finland (Honkavaara
et al 2003), Applanix (Mostafa et al 2003) and the author
(Jacobsen 2000).
The analogue photogrammetric cameras have not been
constructed for the attachment of an IMU causing some of the
stability problems. Also the inner orientation of the analogue
cameras has only the stability required for the classical
orientation. This is different for the new digital cameras like Z/I
DMC, Vexcel Ultracam, Leica ADS40 and Applanix (Emerge)
DSS. The construction has respected from the beginning the
mount of an IMU-system. The line scanner ADS40 cannot be
operated without. Today the use of analogue cameras is
dominating, so most of the investigations are related to them.
The quality of the attitude information has reached a high level,
so with some exceptions the problems of direct sensor
orientation are located more in the position of the projection
centre and the inner orientation. Especially the focal length has
only limited long term stability. Honkavaara et al (2003) are
reporting about a change of a wide angle focal length from
25um to 43um against the calibration certificate. The change of
43um — 25um = 18um corresponds to 0.01% of the flying
height and this is the usual accuracy of photogrammetric point
detection. Cramer (2003) is showing problems of the vertical
component of some long term investigations of Hansa Luftbild.
Baron et al (2003) reports about a change of the focal length of
I5um within 24 days. An explanation for the change of the
focal length can be found at Meier (1978). Theoretical
investigations were leading to changes of the focal length
caused by the vertical temperature gradient in the lens system
based on the cold air outside and moderate temperature in the
aircraft. Discrepancies of the focal length under flight
conditions to the laboratory calibration up to 40pm have been
seen often.
The shift parameters between IMU and the imaging sensor are
changing from day to day. Of course this cannot be explained
by a physical shift of one to each other, but the shift parameters
are strongly correlated with the location of the principal point
and in the flight direction with problems of the time
synchronisation. A separation between the shifts caused by the
GPS positioning and the principal point location is possible if
the reference flight will be flown in two opposite directions.
Due to strong correlation a separation of the principal point
location in the flight direction and a time synchronisation is not
possible, but it is also not required. Cramer (2003) is reporting
about the results achieved by Hansa Luftbild showing a change
of the shift parameters from day to day in the range of 10cm up
to 20cm, but Hansa Luftbild made only reference flights in one
direction, so a separation between the principal point location
and other reasons is not possible. Honkavaara et al (2003) have
identified changes of the principal point location up to 16pm for
the same camera — this is more than the possible
photogrammetric measurement accuracy of objects.
With the exception of the influence to the model set up for a
stereo measurement, the accuracy of the attitude parameters is
usually sufficient. But Dreesen (2001), Honkavaara et al
(2003), Baron et al (2003) and Jacobsen (2000) have seen some
sudden changes of attitude parameters from one day to the other
even after longer time stability. Baron et al (2003) have
identified sudden changes of the attitude relation in the range of
3' and this cannot be neglected for all applications.
7. ACCURACY AND RELIABILITY ASPECTS
If the mentioned problems are respected in the correct manner,
the direct sensor orientation — the determination of the exterior
orientation based on inertial data in combination with relative
kinematic GPS positioning — has reached a high accuracy level
sufficient for most applications. With large scale images, object
point accuracies in the range of 20cm for all coordinate
components can be reached, with a very careful handling even
10cm up to 20cm. This is sufficient for most of the applications.
Still a problem may exist with the set up of models for manual
stereo compilation. Often the rotation yaw or kappa is the week
point causing not acceptable y-parallaxes in the stereo models.
| | x f !
; 317 1 1 Ty f
| P ' |
| Cr = Ee to]
nn | | N ; 20m S y
2 i —t puer
833
Figure 6. y-parallaxes in a stereo model of the OEEPE test
direct sensor orientation integrated sensor orientation
RMSpy-l16um RMSpy=13pm
An extreme case of problems with y-parallaxes in a stereo
model set up by direct sensor orientation can be seen on the left
hand side of figure 6; it corresponds to root mean square errors
of the y-parallaxes of 116 pm (Jacobsen et al 2001). Values
above 20um do cause problems with the stereo impression of
the floating mark. This can be solved by an integrated sensor
orientation based on an adjustment of the exterior orientation
determined by direct sensor orientation and image tie point.
This bundle block adjustment with additional observations does
not require control points. Based on the integrated sensor
orientation the root mean square y-parallaxes of the mentioned
model are reduced to 13pm. In the case of the both OEEPE test
blocks, in the average the RMS y-parallaxes have been in the
range of just 10pm — this is a usual value for model handling.
The integrated sensor orientation requires the determination of
tie points, but today this is a standard procedure solved by
automatic aero triangulation. Of course it improves also the
reliability — the relative relation of the images is controlled by
tie points. For operational blocks usually few check points are
measured to be save for blunders in the data handling. Of
course these check points can be used also for a combined
adjustment which may solve also problems of shift values.
8. CONCLUSION
The direct and integrated sensor orientation reached an
accuracy level sufficient for most applications. An important
aspect is the quite higher flexibility of the block structure like
for traditional block adjustment. Areas with missing object
contrast like water surfaces, desert or forest can be bridged
without problems. Even if few check points are used, their
location can be somewhere within the project area — so special
