ul 2004
cd lever
1 (Cy is
re-eight
r states,
flight is
ered in-
| values
irborne
icult to
methods
1). This
ignment
ibration
- camera
z with a
different
accurate
oves the
je been
oud and
ition are
bundle
together
ientation
> GPS or
centre is
camera
at each
oresight
I] while
data. As
changed
estimate
ariance
e [rad]
:Jations.
A small numerical comparison of three approaches is given
bellow. The test zone consisted of approximately 3x7 image block
and 24 check-points (03cm). The tie points were measured
manually and the AT-GPS aided solution was used as an input to
the ‘2-step’ procedure (with and without time correlation). In
parallel *1-step' boresight determination was calculated. As can
be seen from Table 3, the l-step (1.) and 2-step (II.) estimates
have similar mean values when no temporal correlations are
considered. Both approaches are also too confident in the resulting
accuracy. On the other hand, considering the temporal correlation
in IMU rises the estimate uncertainty that becomes more realistic
for the given type of IMU. However, at the same time, the mean is
closer to the correct value, which in turn increases the mapping
accuracy as shown in the next section and in Table 5.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004
Calibration Flight 1:10000
Boresight Estimation
The higher image quality of a digital camera allows reducing the
scale two times with respect to analogue camera without losing
the details. This fact partially compensates for the smaller format
of the CCD sensor, which requires taking significantly more
photos to cover the same area.
4.2 Mapping Accuracy
The following evaluation will focus on the system absolute
accuracy at discrete points. A test field divided in two areas of
about 25 and 12 GCPs, respectively, will serve the purpose. The
scale of the images that were taken over this test field varies from
1:9000 to 1:11000 and the accuracy of ground control points is at
2cm level. As some GCPs are not specially signalized, the
measurement of their image coordinates may introduce additional
error from 4m to 8um (i.e. 3-8cm in the object space).
T Estimated : ; RMS at GCPs [cm]
Method Estimated MEAN ;
S iol ACCURACY Method Com ams application field
[deg107] GCP Block | oll XY HZ
roll Pitch yaw r D y AT e e 2 4 4
Ei 0.003 1.0311 02421 34313 AT-GPS . 2 2 Ig
I: 1 step 10 12 15
=. i . DQ ^
qose who odas oie sao] s Ap GPS | I: 2step no 9 is 17
time correlation / COIT. ;
«9. : -
UT aep Usine -0.004 | -0.309 | 0235 | 6 | 3| 10 INS | I2 step * 7 i 14
correct correlation time corr.
Table 3: Boresight results according to the employed method
As for the Lidar's boresight, the final procedure has yet not been
finalized but main steps can be briefly outlined:
® Stereoplotting of breaklines on the building roof tops
* Extracting the corresponding lines from the laser points
e Adjusting the plotted and laser-detected lines in each
direction of flight yields the sought boresight angles.
4. SYSTEM PERFORMANCE
The HELIMAP system has undergone several years of experience
in the first two modes of operation (Section 2). Its quality is
appreciated in frequent flying missions related to natural hazards
applications. The functionality and merits of adding ALS became
apparent during a feasibility test that was realized in February
2004. As the evaluation of this data set has yet not been
completed, we turn our focus to CCD based sensor.
4.1 Imagery
The change from analogue to digital camera (Vallet, 2002) was
supported by field and laboratory experiments. Comparisons
between the digital images and the digitized photos revealed that
digital sensors provide sharper and less noisy images than a film-
based imagery as shown in Table 4.
Digital Digitized Film
Transition D ;
B/W A pixel 3-4 pixels
6gsv
Noise (16)
EEE
Table 4: Comparison of digital/analogue photos in terms of
sharpness and noise (gsv grey scale value).
Table 5: Comparison of mapping accuracy between different
approaches to EO determination with an indication of operational
constraints.
The indirect (AT, AT/GPS) and direct (GPS/INS) approaches to
photogrammetric mapping are compared in Table 5 in terms of
empirically estimated accuracy. The direct georeferencing by
GPS/INS is further evaluated with respect to the different methods
of boresight estimation as presented in the previous section. It is
apparent that accounting for temporal correlation in IMU data
during boresight estimate (Table 3) reduces image residuals and
improves accuracy of object coordinates. Although the RMS
values for the direct method are slightly higher than those for the
indirect approach, the demand for providing 20cm-level mapping
accuracy or better is fulfilled. The benefits of direct
georeferencing are, however, numerous, as it avoids many
difficulties that arise when performing automated AT in
mountainous terrain. Adopting this method also considerably
increases the operational flexibility needed in natural disaster
mapping. The AT-GPS approach remains an interesting option for
areas where GCP's are difficult to implement, but the relief and
texture allows successful automation of tie point measurements
procedure. Obviously, the merits of using ALS for fully automated
DTM generation are apparent, but the method is still under
evaluation.
4.3 Cost considerations
The cost of the mapping system is an important and sometimes a
decisive factor for its adoption. Apart from counting the value of
hardware (Table 6), the cost evaluation should also consider the
amount of work related to each mode of system operations (Table
7). As can be seen from Table 7, the image orientation and DTM
generation can rarely be automated in ‘non-standard’ scenarios
involving steep terrain. As these tasks are time consuming, their
liberation by Lidar well justifies the supplementary hardware cost
of USD 35K. The total equipment costs amount to approximately
USD 100'000. This is almost an order of magnitude lower than
