International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
Figure 3: DSS camera head with exoskeleton
(O Applanix 2003).
The dimension of the used CCD matrix is 3.68 x 3.67 cn? (9 x
9 um? individual pixel size) which is less compared to the size
of medium format analogue films (typically between 4.5 x 6 cm?
and 6 x 7 cn). In combination with the two available lens
systems of 55mm (standard) and 35mm focal length (optional)
the resulting field of view is 37deg and 56deg. Comparing the
field of view to the geometry of standard photogrammetric
cameras (23 x 23 cm? format) these values correspond to a
normal-angle (41deg, 30.5em focal length) or medium-angle
(57deg, 21.0cm focal length) image geometry, respectively. The
same situation holds for the base-to-height ratio: With 60%
forward overlap the ratio 9 is 0.42 (35mm focal length) and
0.27 (55mm focal length). Again, both things show the effect of
virtual focal length magnification which will influence the
quality of object point accuracy. This is the reason why the
main application field of the camera is seen in orthomosaicing
followed by photointerpretation or classification (i.e. forestry,
agriculture), or change detection and natural disaster monitoring
and documentation and not in photogrammetric point
determination. In order to illustrate the accuracy potential of the
DSS digital sensor system the results of two different test flights
should be recalled briefly. Both tests were done by the Emerge
production division, where Applanix was responsible for data
processing.
4.1.1 DSS Lakeland test
The Lakeland test flight, flown in December 2002, was mainly
dedicated to evaluate the overall in-flight system calibration and
the potential of photogrammetric point determination —
although this is not the main application field of the DSS
system. The details of the test and the data analysis are already
given in Mostafa (2003). Only a short summary of the main
results is following here. The flight itself consists of 6 flight
strips with standard photogrammetric overlap conditions, i.e.
60% forward and 20% sidelap. Each strip consists of 10 or 11
images, resulting in a total number of 65 images for the block.
Since the flight was done in a flying height of 2000m above
ground using the standard 55mm lenses, the obtained image
scale is about m, = 33000 resulting in a ground sample distance
of approximately 0.3m. The object coordinates of 33
independently determined ground points served as control or
check information to estimate the quality of object point
determination. Two different investigations were done to
evaluate the absolute accuracy of object point determination.
Within the first test the overall system calibration (i.e. in-site
refinement of a priori boresight angles from lab calibration and
control of camera calibration parameters) was performed in the
test area itself. Using the exterior orientations after system
calibration as fixed direct observations (so-called given EO) to
obtain the object coordinates from model-wise forward
intersection is one of the QC/QA features in the Z/I-Imaging
ISAT software that was used in this specific case. Based on the
given EO parameters after calibration the accuracy (RMS) of
object point determination is about 20-25cm for horizontal and
80cm for vertical component, which is in one pixel range for
horizontal and close to 3 pixel in vertical direction. The
maximum deviations (absolute values) arc in the range of
0.55m, 0.80m, 1.90m for east, north, vertical component
respectively. This situation changes when using the given EO
parameters (with certain accuracy) as input data for an
integrated sensor orientation (GPS/inertial assisted AT). Such
approach allows for compensation on small effects in EO
quality or system parameters and results in a better object point
accuracy. The RMS values are considerably smaller indicating a
higher accuracy in object space. The accuracy increase is about
30% compared to the solution based on the fixed given EO
parameters. The maximum absolute deviations are smaller and
reach 0.56m, 0.46m, 1.33m for X-, Y-, Z-components. Still the
vertical component is significantly worse in comparison to the
horizontal component. This is due to the already mentioned
worse base-to-height ratio resulting in less accurate object
height determination.
4.1.2 DSS NASA Stennis Space Centre test
Within the second test briefly cited the quality of the obtained
final orthomosaic was evaluated. Since the orthomosaic is the
final chain in the overall processing flow all different error
sources during processing are controlled which is different from
the first presented test where only the performance of object
point determination was estimated. There seems to be a clear
trend in North-America to perform such final product quality
assessment from independent institutions. In this context the
NASA offers their Stennis Space Center (SSC) facilities for
such system evaluation. This test was done in January 2003,
flown by Emerge production division. Again a 0.3m GSD was
obtained. The final orthomosaic processing was done using the
given NED digital elevation model. For the processing no
additional ground control was used. The result then was
submitted to NASA SSC, where the absolute accuracy
performance was evaluated, by comparing the point coordinates
measured in the rectified orthomosaic to their given reference
values. The results of this test are shown in the Figure 4. The
quality (RMS) about 0.32m is within one pixel, the additional
circles indicate the radii of object point quality within 90%
(CE90) and 9594 (CE95) error radius probabilities. The values
are 0.48m, and 0.55m for CE90 and CE95.
Difference in y (m)
Difference in x (m)
Figure 4: Performance of orthogeneration of DSS imagery
from SSC test (O Applanix 2003).
Ed iua.
a m ERA.
en
rr 05 AN ES —e o r£,»
"to peed pee aay
oF tek
