rs
rs
of
re
in
»ui
ix
le
1d
nd
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
Flights 3 and 4 have been executed about 3 weeks apart, with a
300 mm lens. Their structure is similar to that of flights 1 and 2,
but the 1:18000 scale block is missing, because there was no
practical point in acquiring images at that scale with this lens.
Flight 3 is composed of the 1:5000 and 1:8000 blocks, with the
structure described above. Flight 4 is composed only of the
1:8000 block.
The images (the total number of images is close to 1000) have
been scanned with a pixel size of 14 microns.
1.3 Test field and Ground Control Points
A test field 6 x 4.5 km wide was set up in the city and its
surroundings, which includes either artificial targets (AGCP)
and natural points (NGCP). The GCPs have been measured
with GPS in fast static mode, using three fixed receivers, set up
on vertices of a pre-existing GPS network. To point out and
eliminate possible setup errors, all points have been measured
twice some months later. The estimated inner accuracy of the
network is better than 1.5 cm.
The AGCP set consist of 169 artificial square targets either
metal plates fixed to ground or painted, 35 x 35 cm wide,
homogeneously distributed. The size of the markers has been
chosen in order to be optimal for images in the scale range
1:5000 — 1:8000. Most of the markers have been painted on
paved roads or flat concrete structures.
A smaller set (62 points) of natural GCPs was also available.
1.4 Reference data for calibration
The research unit of the authors of this paper took care of the
measurement and analysis of the block flown at image scale
1:8000 with a 300 mm focal length. The block is made of 11
strips, 7 flown in East-West direction and 4 across the block.
The cross strips are flown twice on the same line but in reverse;
one of the East- West strips is also flown twice. To provide
calibration data and reference data to check the results, a
manual AT was measured. It is planned within the project to
interchange the different flights in order to have independent
calibration and test data. At the time of writing, tough, no other
300 mm block was available, so we decided to used the same
block for both purposes. On the 144 images, 466 tie and pass
points have been measured manually with the software GDS
(Geosoft — Italy). Besides, 199 GCP have also been measured:
154 AGCPs and 45 NGCPs; out of the AGCPs set, 35 points
have been used as control in block adjustment. Therefore,
overall 164 check points were available (figure 1).
The reference system used in the adjustment is a local
tangential frame, with origin at the ellipsoid and y axis in the
meridian plane. The bundle block adjustment and the
calibration of the IMU/GPS system have been performed with
the program Calge of the Politecnico di Milano (Forlani and
Pinto, 1994).
In order to verify the accuracy of the AT and the stability of the
solution, two different configurations for ground control have
been used. In the former, 35 GCP distributed along the block
edges and on the overlap between strips; the latter, with only 4
ground control points at the block corners. No additional
parameters were used.
Fixing the exterior orientation (EO) of the images to the values
computed from the previous adjustment, the coordinates of the
164 available check points (119 artificial, 45 natural) were
determined by forward intersection. Table 2 summarizes the
statistics of the forward intersection.
519
Figure 1 — Plot of the 1:8.000 scale block with location of
control (large pink dots) and check points (red dots)
Block control c, | RMS AGCPs | RMS NGCPs
configuration [um] (cm) (cm)
| X yz IXY zZ
IAT “35 AGCP” 57 123123160155 9313.6
'AT4 “4 GCP” 57 |34|32 |449| 49 |10.7 403
Table 2 - Forward intersection: accuracy of the EO from the AT
measured on the check points.
As expected, also because of the normal focal length, the error
in elevation increases significantly with 4 GCP only, but
accuracy in X,Y is only slightly affected, witnessing the inner
consistency of the block. The estimate for sigma naught is
about half a pixel, which is reasonable for manual
measurements on digital images and is unchanged with less
control. The accuracy of the NGCPs is about half of that
provided by AGCPs.
2. IMU/GPS SYSTEM CALIBRATION
2.1 Introductory remarks
A system calibration is required to account for the spatial
offsets and misalignment between IMU, GPS and camera
frames; moreover, reference to a common time scale is to be
maintained, to allow the interpolation of the IMU/GPS
navigation data to the mid-exposure time of the images.
The spatial offsets IMU-camera can be best obtained by
theodolite measurement, with an accuracy in the order of a
centimeter; an alternative is to include them in the functional
model, perhaps as nuisance parameters, together with the
misalignment angles. Misalignment angles cannot be measured
directly to the necessary accuracy, so they are obtained by
minimizing the differences from the exterior orientation
provided by a standard photogrammetric solution and that
measured by the IMU/GPS sensor.
This calibration problem, which is common also to mobile
mapping system (El-Sheimy, 1996) to our best knowledge has
been first addressed in aerial photogrammetry by the University
of Calgary group (Skaloud et al, 1994). Since, several authors
(Skaloud, 1999; Kruck, 2001; Cramer and Stallman, 2002;
Mostafa, 2002) have proposed variations or alternatives to this
original approach. Calibration techniques can be grouped under
two broad categories, the so called two-steps and one-step
methods (see also Skaloud, 2003 for a review on this specific
