ptable pay-
board soft-
iming code
n of direct
wed to syn-
raw data of
ly estimate
derived by
A 6S’ with
ition by the
raw data of
with a fre-
ver and the
"anon Ixus
.. The self-
the camera
ir viewing.
libration.
y the firm-
to infinity,
eras, could
very short
idy the im-
est field in
In order to
ie lens, the
nera along
analysis of
orientation
t indicate a
1. The cal-
resulted in
5m) image
angle. The
'he control
3 DIRECT GEOREFERENCING
3.1 General
Direct georeferencing is the direct estimation of position and ori-
entation of the camera with sensors on board of the aircraft (i.e.,
without using control points). The position is defined by the three
coordinates of the projection center (Xo, Yo, Zo) in a navigation
frame. The orientation of the camera in the navigation frame
can be described by the three rotation angles roll, pitch and yaw
(ro, Po, yo).
Fig. 3 gives an overview of the available sensors on the UAV and
which sensors can be integrated to obtain position and orientation
of the camera. The interpolation at the exposure-times gives the
position and the orientation of each image.
To improve the height measurements of the GNSS-receiver, the
air pressure sensor was utilised. The WGS84-coordinates were
transformed into the navigation frame with a seven-parameter
transformation. More demanding, than the estimation of the posi-
tion, was the derivation of the orientation from the measurements
of the IMU and the magnetometer, which is described in the fol-
lowing section.
sensors on the UAV interpolation exposure-time
A |
|
PS-recei —| X(t),Y(t) | =
such x Y) a position
Xo. Yo, Zo
air pressure sensor | —» Z(t) um e eum
3 accelerometers |
= cens pr
FI [s ; Dm “Ta | orientation
gyroscopes ae s em To, po, o
=
magnetometer
Figure 3: Direct georeferencing of the images by integration of
all available sensors on board of the UAV.
3.2 Estimation of orientation
The rotation angles r, p, y pl are usually found by integrating the
measured rotation rates wy, wh, ut [^ /s] of the gyroscopes over
time. This works well for high-grade inertial sensors. However,
the inertial sensors on the UAV are based on MEMS technology.
They are small, lightweight, inexpensive, and consume very lit-
tle power. Due to their fabrication process MEMS-sensors have
large bias instabilities and high noise (El-Sheimy, 2009). Thus,
the integration of the angular rates leads to large errors in the ro-
tation angles already after a few seconds. To reduce this errors,
absolute angle measurements are needed. They can be obtained
for roll and pitch from the accelerometers and for yaw from the
magnetometer.
While the UAV is not moving (e.g., when the UAV is hanging
above a defined waypoint) or is just moving very slowly, the three
accelerometers can be used to estimate roll and pitch. On that
condition, the accelerometers measure the three orthogonal com-
ponents g^, 9, 9° ofthe gravitational acceleration g. For the sake
of simplicity, this is shown in Fig. 4 just for the 2D-case. In fact,
in this approach, the accelerometers are used as a tilt meter. Due
to the vibrations on the UAV, before roll and pitch are computed,
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
489
the measured accelerations must be low-pass filtered with a cutoff
frequency of 100 Hz or less.
The time frames where the UAV can assumed to me stable or
constantly moving were determend on the actual observed accel-
eration and rotation values. If for both values only low variations
can be observed the UAV platform can be assumed to be on a
fixed or continous moving position and orientation. Within all
such time frames the accelerometers can be utilised for the drift
compensation.
horizontal plane
p= arcsin 2%
|]
y
Figure 4: By measuring the components of the gravitational ac-
celeration d with the accelerometers, the tilt angles roll and pitch
can be estimated.
The integration of the gyroscopes, the accelerometers, and the
magnetometer for the estimation of the three rotation angles r, p
and y, is shown in Fig. 5. First, starting from the initial values
r(0), p(0) and y(0) the rotation rates w2, w}, W° are integrated.
This gives a first realization of the rotation angles r1, pi, y1. As
described above, the measured accelerations can be used to derive
T2 and p», whereas the magnetometer measures directly yo. At
this point, the rotation angles from the gyroscopes can be com-
pared to the rotation angles from the accelerometers and the mag-
netometer. These differences are the error signals, which can be
used to correct the rotation angles derived from the gyroscopes.
The value of the gain factor k defines how strong the stabilization
of the integral should be. If k is set to 1, the rotation angles are
completely derived by the accelerometers and the magnetometer.
On the other hand, if & is set to O, the rotation angles are com-
pletely derived by the gyroscopes. However, for 0 « k « 1 the
advantages of all sensors can be combined. For the UAV used in
this study, the gain factor k was set to 2 96. It is noted that k need
not be the same for r, p on the one hand, and y on the other hand.
For the presented approach the calculated rotation angles are dom-
inated on short time scales by the measurements of the gyro-
scopes, whereas the accelerometers and the magnetometer cor-
rect the rotation angles over a long time scale.
3.3 Data Streams
The images taken are directly stored on the SD-Card of the cam-
era. The navigation sensor data is downlinked via a Wi.232 con-
nection and stored on a laptop harddisk. In professional kine-
matic multi-sensor systems the GPS PPS signal is used for the
synchronization of all data streams, which is, however, not avail-
able for the described components.
The two streams of navigation data, on the one hand, and im-
ages, on the other hand, are synchronized via a signal generated
