The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008
92
200 400 600 800 1000 200 400 600 800 1000
(a) (b)
Figure 5. The radial systematic lens distortion in the image
from calibration report (a) and in-situ calibration (b)
(pm)
24 u m
Figure 6. Systematic image errors determined by introduced
additional parameters to the reference adjustment
4. BORESIGHT MISALIGNMENT
The image orientation determined by GPS-supported bundle
block adjustment, with improved image coordinate and focal
length in approach 5 was used for the determination of the
boresight misalignment. The orientations from the GPS-
supported bundle block adjustment were compared to the
orientations obtained from GPS/IMU processing. The shift
parameters were determined by comparing the projection
centers from the reference adjustment to the GPS/IMU-derived
projection centers as well as to lever arm measurements.
In general, the IMU is fixed to the camera body as close as
possible and is aligned parallel to the camera. In our installation,
the camera was installed perpendicular to the flight direction,
resulting in a 90° rotation between the x axis of camera and that
of the IMU. The orientations from the reference bundle block
adjustment were transferred into roll, pitch and yaw and the
boresight angles were determined as 0.18476° for roll, 1.29884°
for pitch and 0.34447°for yaw. The determined shift values
were -0.384 m for X, 0.076 m for Y and 0.050 m for Z.
The boresight angles were also estimated using the POSCal
utility of the Applanix POSEO software version 4.1. The
POSCal computation is based on least squares adjustment and
the required inputs are the image coordinates, control points and
GPS/IMU-derived projection centers and orientations. The 90°
rotation between the x axes of the camera and IMU had to be
also defined for the computation. The determined boresight
angles were 0.16867° for roll, 1.27303° for pitch and 0.40910°
for yaw.
The GPS/IMU-derived attitudes and positions were improved
by the BLUH and Applanix POSCal boresight misalignment.
Based on the improved GPS/IMU derived attitudes and
positions, the object coordinates of measured tie points and
check points were computed by combined intersection (direct
sensor orientation). The 25 control points derived from the
LiDAR point cloud were used as check points. The a 0 of the
direct georeferencing and root means square errors at check
points can be seen in Table 2.
Approach
Of)
[ft
m]
RMS at Control
Points [m]
X
Y
Z
1
Direct georeferencing using
BLUH boresight misalignment
52.2
1.1
4
0.
79
5.
35
2
Direct georeferencing using
Applanix boresight misalignment
46.0
1.1
8
0.
70
5.
57
Table 2. Direct georeferencing results in UTM
The effect of the orientation discrepancies can be seen as y
parallaxes. The y parallax in the model is important for stereo
model setup. The comparison of the model y parallaxes from
direct sensor orientation based on BLUH and Applanix
boresight angles, and GPS supported bundle block adjustment
can be seen in Figure 7. The similar results obtained for both
direct sensor orientations clearly indicate an unacceptable
quality for stereo models.
Figure 7. Comparison of the y parallaxes
To further check the quality of the sensor calibration and
boresight misalignment, orthoimages were generated using GPS
supported bundle block adjustment results and GPS/IMU
derived attitudes and positions improved by BLUH boresight
misalignment. The effect of orientation discrepancies to the
orthoimages can be seen comparing the LiDAR intensity
images and the generated orthoimages in Figure 8. The in-situ
camera and boresight calibration were determined based on the
data collected on May 25, 2005. The performance of the in-situ