overview of the different approaches most commonly used for the 
orientation of the image data. 
Sensor- 
type 
Approach 
indirect 
direct 
frame 
Aerial Triangu 
lation (AT) 
GPS/INS- 
Integration 
line 
AT + kine 
matic model 
GPS/INS- 
Integration 
Table 1: Different orientation approaches for imagery 
2.1 Indirect method 
In classical photogrammetry using full frame imagery (analogue or 
digital) the georeferencing problem is solved indirect using ground 
control and applying geometric constraints between image points 
and object points. For single image data this procedure is done 
by spatial resection, which can be generalized to an aerial triangu 
lation (AT) for multiple images. Within this adjustment the pho- 
togrammetric collinearity and coplanarity equations are used to 
connect neighbouring images via tie points and to relate the local 
model coordinates to the global coordinate system. The exterior 
orientation parameters for the perspective centre of each image 
are estimated as one group of the unknown parameters in the ad 
justment. 
In principle this approach can be transferred directly from frame 
to line imagery acquired by a digital push-broom scanning sys 
tem. For push-broom systems each image consists of one line 
in general, using multi-line scanners two or more image lines are 
recorded simultaneously, therefore the image consists of several 
image lines (e.g. within the stereo module of the DPA system three 
pan-chromatic CCD lines are used for data recording). Compared 
to frame sensors the image geometry of line scanners is much 
weaker and the orientation parameters have to be reconstructed 
for each line image. Assuming a line scanner with a data rate of 
200Hz yields in 1200 unknowns within one second for the position 
(Xo, Vo, Z 0 ) and attitudes (tv, <p, k) of the camera. However, there 
is not enough information available to estimate this large amount 
of unknowns in an adjustment procedure. Therefore, the exterior 
orientations are determined explicitly for distinct points of time only, 
the so-called orientation points. The trajectory of the sensor dur 
ing the time intervals between these points is interpolated using 
an appropriate kinematic model for the sensor platform (AT + kine 
matic model). This approach reduces the numbers of unknowns 
significantly and can be applied very well for space borne sensors 
(Kornus et al., 1998). 
Due to the high dynamics of an airborne environment the system 
has to be expanded with an INS for the measurement of the short 
term movements. Using a kinematic model is replaced by an INS 
that measures the position and attitude data for each image line di 
rectly. Although the INS now provides direct measurements of the 
sensor orientations and these data are introduced in the adjust 
ment, the orientation determination is mainly based on the pho- 
togrammetric constraints used to determine the orientation points 
and therefore this approach still belongs to the indirect methods of 
image orientation. For airborne scanning systems the potential of 
this approach is shown e.g. in (Hofmann et al., 1993), (Heipke et 
al., 1994). 
2.2 Direct method 
First attempts of direct measurement of exterior orientation in the 
field of photogrammetry were done since the early thirties of this 
century. Driving force of these investigations was the aim to sig 
nificantly reduce the need of ground control. At that time most of 
these attempts were limited due to their accuracies and the lack of 
operationality. 
With the advent of the global satellite navigation systems (e.g. 
GPS) and the reduced costs of inertial navigation systems (INS) 
this situation changed tremendously. GPS offers the possibility 
to determine position and velocity informations at a very high ab 
solute accuracy. The accuracy level is dependent on the pro 
cessing approach (absolute, differential), the used type of observ 
ables (pseudorange-, doppler-, phase-measurements) and the ac 
tual satellite geometry. To obtain highest accuracy the differential 
phase observations are used. Solving the ambiguities correctly 
and assuming a reasonable satellite geometry, positioning accura 
cies up to 10cm are possible for airborne kinematic environments 
with remote-master receiver separation below 30km. Typical accu 
racies for the velocity determination are at the level of a few cm/s 
(Cannon, 1994). 
The principle of inertial navigation is based on the measurements 
of linear accelerations and rotational rate increments of a body rel 
ative to an inertial coordinate frame. The actual position, velocity 
and attitude informations are obtained from an integration process. 
Starting with an initial alignment to get the initial position, velocity 
and attitudes, the first integration of the angular rates and linear 
accelerations gives attitude and velocity information. After a sec 
ond integration step the position informations are available. Due 
to these integrations the accuracies of INS are not constant but 
time dependent. Due to the quality of the used inertial sensors, 
the accuracy is very high for short time spans but degrades with 
time caused by accumulating errors within the integration process. 
Additional errors are introduced from errors in the initial alignment. 
To reduce the systematic errors the INS has to be supported by 
additional data. In the high dynamic airborne environment only 
GPS can meet these requirements, therefore GPS is an ideal sen 
sor for integration with inertial data. Due to the complementary 
error behaviour, the high long term stability of the GPS measure 
ments can be used for bounding the growing INS errors. Tradi 
tionally, this GPS/INS integration is realized in a Kalman filtering 
approach. Within this process the GPS position and velocity infor 
mation is used to determine the errors of the chosen error states. 
For medium to high quality INS a 15-state error model, consisting 
of 9 navigation errors (position, velocity, attitude) and 6 sensor spe 
cific error terms (gyro drift, accelerometer bias) can be sufficient for 
many cases (Skaloud and Schwarz, 1998). Additional error terms 
can be introduced due to the physical offsets between the GPS 
antenna and the INS. 
Several tests have shown the high potential of these integrated 
GPS/INS systems for georeferencing of image data. Especially in 
the last years, these systems have been tested extensively for the 
orientation of airborne analogue or digital frame cameras as well 
as for digital line scanners (table 1). Comparing the exterior orien 
tations from GPS/INS with reference values from aerial triangula 
tion, accuracies of camera positions of 10-15cm and attitude accu 
racies up to 15 arc sec were achieved for the measured orientation 
parameters. Using the position and attitudes directly measured 
from GPS/INS for the orientation of a standard photogrammetric 
wide-angle camera and recalculating image points by spatial for 
ward intersection of image rays, accuracies on the ground of less 
than 20cm for the horizontal and less than 30cm for the vertical 
coordinates could be obtained from a flying height of 2000m above 
ground (Wewel et al., 1998), (Hutton and Lithopoulos, 1998). 
The positions and orientations from GPS/INS do not refer to the 
perspective centre of the imaging sensor directly. Caused by trans 
lational and rotational offsets, the GPS antenna and the centre of 
the inertial system are displaced from the camera. Additionally, 
the attitudes from GPS/INS are calculated from the rotation of the 
INS body frame defined by the INS sensor axes to the local level 
frame. The INS axes do not coincide with the image coordinate 
frame. These offsets have to be taken into account before ap 
plying the orientations for the georeferencing of the imagery. The 
translational offsets are determined using conventional terrestrial 
survey methods after installation of the system in the aircraft used