accelerometers gyroscopes magnetometer
a5, a5, ab [m/s?] «P, ub, wb [ /s] y[°]
ERA r(0)
low-pass
EN J * 00
SE SAI y(0)
T2, P2
from ge
T2, P2 T1,P1 yi y2
FON EN
Es (=) 5
NS L No /
Figure 5: Estimation of roll r, pitch p, and yaw y
at each triggering of the camera, which is fed into the downlink
stream. The time lag between exposure and insertion of this sig-
nal in the downlink stream was determined once by a photograph
of the laptop screen were the downlinked navigation data stream
was displayed in real time on the screen. Based on this simple
procedure a time lag of 0.75s + 0.11 s could be estimated. This
solution became possible with some minor adaptions of the open
source MikroKopter flight control firmware.
4 EVALUATION
The quality of the exterior orientation computed by the process
presented in section 3 can be determined in different ways. The
navigation sensors can be mounted, independently of the UAV
system with the camera on a different platform which is equipped
with a high end position and orientation system. This provides
reference data, which is directly of the same type as the acquired
(low cost) sensor data. However, this is not the ordinary operating
principle of the UAV. As the aim is the direct georeferencing of
images, we compared it to position and orientation values deter-
mined by indirect georeferencing, obtained from a bundle block
adjustment.
The exterior orientations of a bundle block adjustment can act
as an absolute reference, if systematic errors in the bundle block
triangulation can be ruled out. Typical sources for that would
be insufficient camera calibration or errors in the control points.
The camera calibration was described above. Control points were
measured using RTK GPS, providing an absolute accuracy (1
sigma) of ~ 2cm, with a relative accuracy below 2cm.
A test fligh was performed over a model aircraft field. Fig. 6
shows an orthophoto mosaic, created from the acuired imagery.
A layout of 4x6 control points (circular shape, see also Fig. 2)
was used for the test of direct geo-referencing.
The flight (cf. Figure 7) was organized along a predefined route,
with waypoints. At each waypoint, the UAV remained stable for
~ 10 seconds (user defined). Images were taken at a constant in-
terval of 5s and the focus was set to infinity. The flight lasted
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
490
Figure 6: Test area with 24 control points on a model aircraft
field.
flight time 412s
number of direct georeferenced photos 72
number of indirect georeferenced photos 53
trajectory AX 34.7 m
trajectory AY 27.4m
max. flight height above ground 25.5m
exposure time of the photos 1/1000 s
Table 1: Flight statistics.
188 take off
e landing
s » photos
168
19640
158
S
v
p
z 950
x 19590
aov [m]
n,.
M^
oY
Figure 7: Trajektory of the waypoint flight. Due to a miscalibra-
tion of the magnetometer the waypoint flight was stopped and the
final flight and landing was performed manually.
for approximately seven minutes, and reached a height of 25m
above ground. Due to a problem with the calibration of the mag-
netometer (new firmware version) the automated waypoint flight
was interrupted and the following flight path (zz 50 96) and land-
ing was performed by manual remote control. Additional flight
parameters are given in Tab. 1.
The aerial triangulation was performed using the program pack-
age Orient/Orpheus developed at the Vienna University of Tech-
nology. The adjustment led to a mean accuracy (1 sigma) of the
image positions of 2.2cm, 2.6cm and 0.8 cm for the x-,y- and
z-coordinate, respectively. The mean accuracy (1 sigma) of the
rotation angles r, p, y of all adjusted images are 0.075 ^, 0.064 2
and 0.015 °. The mean accuracy of the x,y, and z control point
co-ordinates are 0.2 cm, 0.2 cm and 0.7 cm, respectively.
The position and orientation (direct georeferencing) was com-
puted according to the description in Sec. 3. However, at the
time of processing the determination of the yaw angle was just
based on the magnetometer observations. The resulting values
