TYPE
consists of a
(Litton LTN
System was
ectrographic
Ltd for this
1 pushbroom
travels along
different set
scanline, and
ovements on
wn in Figure
, Xp: Jp Ky,
i drb were
> area flown,
e orientation
b
[; dR will
alled in the
f the camera
>. However,
ree or more
k is defined
enience four
djustment is
& Buchheit
, (1992) and
e taken into
EM). The
is currently
7.1 Calgary Test Flights
A special variant of the casi was developed by Itres
Research Ltd. for this project. This sensor contains fore,
nadir and aft look directions giving it stereo capabilities.
The sensor has 512 pixels along each scanline and a field of
view of 43.5°. The fore and aft look directions are 16° off
nadir. These limited off-nadir angles were due to the CCD
geometry.
Three flightlines were flown over the city of Calgary. Each
flight line was 10,000 lines long with an along track pixel
size of 1.6 metres and an across track pixel size of 2.5
metres. Included on board the aircraft were the casi sensor,
a Litton LTN-90 inertial system and an Ashtech P12
receiver. Another GPS receiver was placed near the centre of
the test area. GPS positions were processed to 0.50 metre
accuracy.
Due to the fairly large pixel size, large area and urban nature
of the area, placing targets on the ground was impractical.
Therefore, road intersection centres were used as targets
instead. These control points are somewhat poor due to
errors in matching the road intersection centre on the ground
to the image. However, in a production environment, this
would be the most likely way control would be established.
7.2 Test Results
To test the accuracy of the system, bundle adjustment results
are shown. These would be equivalent to geocorrection
accuracies if the DEM was taken into account. Each
: : b ; :
adjustment solves for a different dR, for each flight line.
The values for the orientation differences &, ® and x are
compared for each flight line.
Three different solutions are shown for the above mentioned
data set. They are:
a) First, the bundle adjustment was run with all control
points and fore, aft and nadir look directions. This
gives the optimal solution from all available data and is
used to compare results.
b) Next, the bundle adjustment was run with only 4 control
points and fore, aft and nadir look directions. This
corresponds to the case of minimum ground control for
calibration. The remaining control points are included
as check points. This would be a standard stereo
Scenario.
c) Lastly, the bundle adjustment was run with 4 control
points and only the nadir look direction. The
remaining control points are included as check points.
This scenario is would be used with the standard multi-
spectral casi.
Test (a) results show that the overall horizontal accuracy is
about 1.2 pixels (3.0 metres). The largest control point
error was 5 metres (horizontal). Larger errors in the Z axis
are due partially to a poorer height geometry as seen by the
Standard errors, and also due to poor control point
registration between the image and ground.
Control points. 2.778 2.861 14.88 27
RMS (m)
Check points
RMS (m)
Image residuals
(pixels)
Avg. standard 2.32 2.11 8.60 281
error (m)
Table 6: Test (a) Results, All Control Points
Used With Stereo Pushbroom Sensor
With improved targeting, the control point residuals will be
reduced as they are currently the largest error source. Other
error sources result from the GPS and INS, but these are both
below the sub-pixel level. There are also errors due to the
calibration, but they are difficult to quantify individually.
The principal point (=CCD centre) is most difficult to
calibrate. However, erroneously calibrated parameters will
be partially compensated by the INS/sensor offset angles.
0 (deg.) 0 (deg.) K (deg.)
Flightline 1 1.9697 5.7278 3.1517
Flightline 2 1.9203 5.5642 3.4284
Flightline 3 1.9998 5.7948 2.7574
Table 7: INS/Sensor Offset Angles for Test (a)
The values for the INS/imaging sensor orientation offset
b
dR, are shown in Table 7. Although they are relatively
consistent, there are some variations present. Most
significant variations are seen in the K (yaw) values. This
larger variation is expected because the system, with its
fairly narrow swath angle (43.5°), is less sensitive to yaw.
Other differences may also be due to the fact that flightlines
1 and 3 were in one direction and flight line 2 in the
opposite direction.
X Y Z no.
Control points. 1.497 1.401 2.325 4
RMS (m)
Check points 2.646 3.701 22.33 23
RMS (m)
Image residuals 0.99 0.72 1057
(pixels)
Avg. standard 2.43 2.18 301
error (m)
Table 8: Test (b )Results, 4 Control Points Used
With Stereo Pushbroom Sensor
Table 8 shows results of test (b) using only 4 control points.
These results are very similar to those of Table 6 which
means there is no significant accuracy degradation due to
using fewer control points. Four points are sufficient for
this size of block. This means that after calibration, the
georeferencing system performs with the expected accuracy.
Tables 9 and 10 show results using only the nadir channel.
This is the geometry of the standard casi sensor. The errors
in y are slightly larger than from the (a )and (b )tests.
199