can be considered as the precision of the trajectory from direct
georeferencing. The higher difference in X is attributed to prob-
lems in the synchronization between the two postion streams.
(Blaha et al., 2011) extend the above investigations by using dif-
ferential post-processing of GPS for a Falcon 8 UAV. The GPS
receiver used is a ublox LEA 5T. The precision, determined as
before, is 35 cm, 28 cm, and 38 cm. The mean value of the differ-
ences is in the order of 40 cm.
(Haala et al., 2011) compare the results of automatic aerotrian-
gulation to the direct georeferencing positions for a fixed-wing
UAV. Their platform is equipped with a Locosys LS20031 GPS
receiver and an air pressure sensor, which gives the Z-coordinate.
Their experimental results of a std.dev. of 3 m in each coordinate
confirmed their expectations.
(Eugster and Nebiker, 2008) describe direct georeferencing of a
small UAV in an application context and estimate a-priori the
error in object space due to direct georeferencing. For a flying
height of 25 m above ground this results in a error due to roll and
pitch of 50 cm, due to yaw of 1 m, in ‘position’ of 3 m. They con-
clude that the overall accuracy of 6 m to 15 m for flying heights
of up to 300 m are sufficient for many applications, with the ad-
vantage of providing real time geo-referencing.
2 THE UAV
For this research a low-cost multicopter with four rotors based on
the project ‘MikroKopter’! was used (Fig. 1), see also (Briese
and Glira, 2011).
SRE p i & Diu
Figure 1: UAV system based on the ‘MikroKopter’ project.
Multicopters usually have four (quadrocopter), six (hexacopter)
or eight rotors (octocopter), which are arranged in a horizontal
plane. For continually levelling the aircraft, a processing unit es-
timates the inclination by evaluating the measurements of an IMU
(Inertial Measurement Unit) consisting of three gyroscopes and
three accelerometers. Next to the IMU some multicopter systems
have further sensors on board. These are usually a magnetometer,
an air pressure sensor, and a GNSS-receiver. Due to the use of all
these sensors, multicopter systems typically behave very stable
in the air and support the human operator to a high degree. Since
these aircrafts became quite affordable over the last years, their
use in research and commercial applications is currently increas-
ing significantly.
lyww.mikrokopter. com
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
The ‘MikroKopter’ costs about 2000 $, offers an acceptable pay-
load (250 g) for use in photogrammetry, and the on board soft-
ware is open source. The availability of the programming code
of the main processing unit was essential for the aim of direct
georeferencing. Modifications of the source code allowed to syn-
chronize the camera with the sensors and to save the raw data of
these sensors. The sensor data was used to continually estimate
position and orientation of the UAV. The position was derived by
the measurements of the GNSS-receiver (‘u-blox LEA 6S’ with
SBAS/EGNOS) and the air pressure sensor, the orientation by the
measurement of the IMU and the magnetometer. The raw data of
the IMU and the air pressure sensor can be recorded with a fre-
quency of 20 Hz, whereas the data of the GNSS-receiver and the
magnetometer is only available at a frequency of 2 Hz.
For image acquisition a digital compact camera (‘Canon Ixus
80 IS’) with a sensor resolution of 8 MP was chosen. The self-
built camera mount is vibration-damped and connects the camera
rigidly to the body of the UAV. It is restricted to nadir viewing.
This results in an unknown, but constant mounting calibration.
Using the ‘Canon Hacker Development Kit” to modify the firm-
ware of the camera, the focus could be set permanently to infinity,
and automatic shut down, as typical for consumer cameras, could
be turned off. In order to reduce the image motion, a very short
exposure time had to be defined (Fig. 2). For this study the im-
ages were taken in a constant time interval of 5 s.
The camera was calibrated with a photogrammetric test field in
a laboratory with a large number of control points. In order to
study the influence of gravity on the sensor and/or the lens, the
images for calibration were taken by rotating the camera along
the horizontal pointing viewing axis. The subsequent analysis of
the bundle block adjustment with free inner and outer orientation
parameters for each individual camera rotation did not indicate a
significant influence of gravity for the utilised camera. The cal-
ibration procedure of all acquired calibration images resulted in
residual errors (1 sigma) of = 0.2 px.
Figure 2: Example of an UAV (flying height: approx. 25m) image
with an exposure time of 1/1000 s. Notice the low sun angle. The
insert zooms to a control point used for evaluation. The control
point ID is printed on an A4 paper.
2http://chdk.wikia.com/wiki/CHDK
