Paul Pope
4. DISCUSSION
The numerical modeling test results indicate that the PMIIM method has the ability to correct an error prone trajectory
so that it can be used to georeference airborne scanner imagery. Using the error prone trajectory without correction (the
"do nothing scenario") yields a planimetric RMSE of approximately 34 meters. Note that this is consistent with the 3
GSDs (30 meters) of noise introduced to the actual trajectory to form the measured trajectory. Using the PMIIM
method to correct the trajectory, the planimetric RMSE was reduced to approximately 21 meters. Taking advantage of
the coupling between the roll and pitch with the other exterior orientation parameters reduced the compute time from
approximately 6 hours on a 600 MHz machine to only 30 minutes on a 200 MHz machine. However, the RMSE was
not reduced significantly (32 meters), though the minimum and maximum errors did decrease. This is most likely due
to the fact that corrections to the pitch cannot be ignored in areas of significant terrain relief.
The spatial location of the largest planimetric errors occurs near the edges of the scan in areas of high relief (ie,
occultation problems), and where there is large under and over-sampling. Under and over-sampling are excessive in
this numerical simulation, as is the level of noise compared to the amplitudes of the position exterior orientation
parameters. Imagery from an actual flight usually contains much less of these effects. Also, most scanning systems
implement roll correction, which would most likely eliminate the need to correct the trajectory for that exterior
orientation parameter. Another adverse effect maybe that the spatial autocorrelation factor only represents an average
autocorrelation lag length. There are parts of the image for which the lag is zero (i.e. every scan line is not significantly
correlated). These facts may explain why there are sporadic mismatches between the corrected trajectory and the actual
trajectory (Figure 2). This in turn led to a reduction in planimetric RMSE which was only equal to about 1 GSD,
Finally, we expected that the planimetric error in the cross-track direction is usually less than that in the along-track
direction since there is an approximation to perspective viewing geometry in the cross-track direction, as opposed to the
parallel viewing geometry in the along-track direction. However, it is interesting to note that in this study the
planimetric error in X (Easting) is larger than the planimetric error in Y (Northing), since the cross-track and
along-track scanning directions are nominally associated with the X and Y directions, respectively.
This research also reveals one of the major impediments to using a parametric method. That is, the labor involved in
creating the inputs to such a method. This is apparent from the number of pre-processing steps described previously.
Also, this method will probably find its greatest utility when applied to imagery from an airborne scanner survey in
which no trajectory information has been recorded (e.g. historical imagery), or for which only relatively inaccurate and
temporally coarse trajectory information is ava.ilable. This method most likely would not be able to enhance the
trajectory accuracy available from highly accurate and temporally fine trajectory measurements, such as those available
from INS/GPS systems.
5. FUTURE WORK
This research has investigated the initial testing of a new method to georeference airborne scanner imagery. Althougha
successful proof-of-concept of the PMIIM method has been accomplished, other numerical simulations would be
interesting. Testing of the PMIIM method's sensitivity to variations in spatial autocorrelation and trajectory noise could
address some of the concerns expressed in discussing the results of these experiments. Some questions which might be
addressed in the future are "Will reducing the spatial autocorrelation factor lead to better results?", "Can similar results
be obtained through smoothing the input trajectory?", and "What is the result of using only every 18th record of the
input trajectory and splining in between records?". However, more applications oriented results can be obtained by
applying the PMIIM method to the problem of georeferencing actual airborne scanner imagery. The authors ar
presently engaged in applying the PMIIM method to real ATLAS imagery. Future work will describe the results of
using the PMIIM method to georeference real imagery, provide a more in-depth accuracy assessment and statistical
analysis, and a comparison with traditional techniques.
ACKNOWLEDGEMENTS
The authors wish to extend their sincere thanks to the following organizations. To the Wisconsin Department of
Natural Resources for their contribution of digital orthophotography to this project. To Microlmages, Inc. for suppor
of Mr. Pope’s dissertation work. To the Environmental Remote Sensing Center at the University of Wisconsin for
logistical support. To the Commercial Remote Sensing Program at NASA's Stennis Space Center for financial and
technical support. And to the Los Alamos National Laboratory for computer support, and especially to the LANL
Graduate Research Assistant program for enabling Mr. Pope to pursue his educational goal.
738 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
DI
The
by
Bre
Che
Eth
Ji,
Loc
Mc
Mc
Me:
(/
Nat