914
December 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
systems for each exposure center, these are not survey quali-
ty. Two control points alone placed the model on a single axis
about which the model was rotated by the algorithms in or-
der to achieve the best results. This can be seen most obvious-
ly in the Sony 200ft 2 GCP difference grid. The deviation is
similar to the lidar only in a streak through the center of the
project, forming a line tracing the path of the ground control.
The other ends of the project are either above or below the ac-
tual terrain. The Canon 400ft flight had a similar but unique
error pattern, with the center of the model showing low de-
viation and the outside edges showing very high deviation.
This could indicate systematic errors present in the camera
calibration, and the error could possibly be due to high radial
distortion causing warping of the model. More testing needs
to be done here, but the issue became less apparent as more
control was added.
The addition of a third ground control point at minimum
seems to be the best practice. However, five ground control
points consistently produced the best results with the UAS cam-
eras. There was a small increase in error (less than 2cm) when
using 5 ground control points with the UCFp datasets. The
team attributed this to compounding error between the onboard
GPS/IMU sensors and the existent error in the ground control
(<3 cm). Over a larger area or with higher accuracy ground con-
trol points, it is likely that a result with a consistent RMSE of
around 1 to 2 cm would be possible. It is interesting that while
the UltraCam Falcon Prime performed the best under all condi-
tions, the smaller sensors were able to nearly match the results
when five ground control point are employed. The accuracies
achieved here may be more than suitable for a large number of
applications, reinforcing the use of UAS and non-metric camer-
as as a legitimate tool in many situations.
Lessons Learned
It is clear that the Canon 22MP camera mounted on the fixed
wing aircraft performed better than the Sony A7R 36MP
camera on the rotorcraft in lower control situations. How-
ever, with more control, the Sony sensor produced accura-
cies comparable to the UltraCam Falcon Prime. Using the
AT solution as a guide, its seems the root cause may be in
the stability of the lens. Despite the high quality Zeiss 35mm
lens on the Sony A7R, the adjustments performed by the soft-
ware would suggest an instability in the lens, manifesting in
poorer correlation, especially in Z. Whereas the Canon cam-
era lens system has been modified by the UAS manufacturer,
resulting in a more stable lens system and more predictable
results. In the case of the Zeiss lens, the situation of less con-
trol caused the errors to manifest without the ability to be
systematically corrected.
In aerial triangulation, each camera performed better at
the larger pixel sizes, especially the smaller format sensors,
suggesting that the increase in the image amount compound-
ed the level of tie point uncertainty between control points
and reduced the overall quality. It is also possible the very
high resolution images resulted in a more homogenious scene
with fewer ground features to be matched by the algorithm,
resulting in more false-positive tie points, i.e., artifacts. While
this was the case in triangulation, the higher resolution im-
agery resulted in better residuals in Z when comparing the
check points to both bare earth and surface model products.
This suggests a consideration of the end product must be un-
dertaken before a survey is conducted. The user must weigh
the advantages of a higher relative AT accuracy with lower
resolution data versus higher resolution, rich imagery and
high absolute accuracy in Z. This will largely be decided by
the final data application.
Continued Development
Keystone is actively integrating and testing IMU and GPS/
GNSS solutions that could further reduce the number of
GCPs necessary in all but the highest accuracy requirement
products. The use of high quality GNSS has greatly changed
the landscape of traditional photogrammetry from large for-
mat cameras and will most certainly do the same for UAS
imagery. Adding the high frequency data collection of an
IMU to the solution will further the algorithms possible in
post-production and in the bundle adjustment/self-calibra-
tion routines.
Conclusion
Keystone found that with the proper control of a UAS solu-
tion a quality product can be generated for almost any solu-
tion. Despite pricing considerations and the effects of many
more images per acre, the UAS imagery solution seems to be
a viable solution in many circumstances. This is especially
true when tools that will reduce the reliance on GCPs become
available and mapping sensors are purposely built for UAS
imagery capture.
References
Tucker, C.J. (1979). Red and photographic infrared linear
combinations for monitoring vegetation. Remote sensing
of Environment, 8(2), 127-150. Retrieved from
Federal Aviation Administration. (2016).
FAA Aerospace
Forecast: Fiscal Years 2016-2036.
FAA. Retrieved from
McCarthy, N. (2015, 10 19).
Forbes Business
. Retrieved
from
Forbes:
Shen, L. (2016, 5 25).
Drone Sales Have Tripled in the
Last Year
. Retrieved from Fortune:
