PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2016
913
Sony producing height and XY accuracies comparable to the
UltraCam Falcon.
Due to the poor results in RMSE from the height measure-
ments from the Sony and Altavian tests at 200 ft. with only 2
or 3 GCPs, Keystone used another software package to deter-
mine if better results could be achieved. SimActive Correla-
tor 3D was used to process the Sony A7r imagery with 3 and
5 GCPs. The result for the 3 GCP test was vastly superior
and produced a solid solution. However, the result from the
5 GCPs test demonstrated only slight improvement and the
RMSE in Z was actually inferior to the Pix4D Mapper Pro re-
siduals. This clearly shows the differences in the algorithms
of these two packages create different results with the same
GCP and image combinations. Further study
should be made to determine if each of the many
software packages would deliver such varied re-
sults when processing the same sets of imagery.
The results are compared in the tables below:
Elevation Models
Keystone utilized Teledyne Optech’s LiDAR Map-
ping Suite Pro to produce refined and geo-reg-
istered LAS files from the acquired lidar data.
Similar LAS files were then produced from the
imagery datasets using Pix4D Mapper Pro for the
UAS data, and Trimble’s Inpho Match-T for the
UCFp data. The photogrammetric image match-
ing produced point clouds with very high densi-
ties. For example, the Sony platform produced a
point density exceeding 1,300 points per square
meter. Terrasolid’s TerraScan software was used
to compare the ground control points to the un-
classified point cloud surface models. Unfortu-
nately, the ground targets set at this site were
not visible in the lidar intensity data. Therefore,
checkpoint data could only be used to validate the
vertical accuracy of the lidar data.
To compare the functional accuracy of each ele-
vation model, the derivative bare earth digital el-
evation models were tested. All photogrammetric
and lidar point clouds were classified using Ter-
raScan, with points being identified and placed
into either the ground or non-ground class. Using
the ground class, a triangulated irregular net-
Table 4: Comparison of Correlator 3D and Pix4D Mapper Pro results on Sony
A7r 200 ft. Dataset.
RMSE with 3 GCPS (cm)
RMSE with 5 GCP (cm)
X
Y
Z
X
Y
Z
Correlator 3D
2.09 2.76 4.64 2.23 2.29 4.49
Pix4D Mapper Pro
2.72 2.03 20.69 2.70 2.30 3.94
work (TIN) elevation model was interpolated from
the points. These TINs were then gridded to an ele-
vation raster grid with a cell size of 0.5 meters. The
lidar and photogrammetric DEMs were then com-
pared to the checkpoints using QT Modeler. Results
from the DEM are in Figure 6 and point cloud control
reporting can be seen in Figure 5.
S
ummary
A number of conclusions can be drawn from these results.
Perhaps most strikingly, the introduction of a third ground
control point dramatically increases the reliability of the re-
sults from the low altitude Sony and the high altitude Canon
when compared to the 2 GCP flight. This was reflected in the
AT results and the implications of these high residuals are
apparent in the product. This is of no particular surprise; nei-
ther the Mavrik nor Altavian platforms have high accuracy
GPS/IMU sensors onboard, as is the case with the UCFp and
Galaxy lidar. While it was possible to use the telemetry data
recorded during flight to determine photo locations from the
Figure 5: RMSE results on check points by sensor of Point Clouds in TerraScan.
Figure 6: RMSE results on check points by sensor of DEM in QT Modeler.
