for the photo flights. The rotational offsets between the INS sen 
sor axes and the camera coordinate system cannot be observed 
via conventional survey methods. Therefore, these rotational off 
set or misalignment angles between the INS and camera system 
have to be determined with in-flight calibration using a small num 
ber of tie and control points. Nevertheless, if there are no relative 
movements between the different sensor components, these off 
sets should remain constant for several survey campaigns. There 
is some ongoing work to prove the stability of these displacements 
over a longer period of time. 
The quality of the integrated GPS/INS positions and attitudes is 
highly correlated with the quality of the updating information from 
GPS. Even though the INS informations can be used to bridge 
short GPS outages or to detect small cycle slips of the carrier 
phase measurements, the overall performance will degrade if the 
GPS position and velocity update informations are of minor quality 
for a longer time interval. The inertial data can only be used to 
detect GPS short term failures. The correction of long term sys 
tematic errors is not possible. Especially in case of photogram- 
metric applications where the distance between remote and mas 
ter receiver can be very large due to operational reasons, at least 
constant offsets for GPS positions have to be expected resulting 
from insufficient modeling of the atmospheric errors. Additionally, 
errors might be introduced from incorrect datum parameters for 
datum shift, remaining systematic effects from the imaging sensor 
or - quite simple - erroneous reference coordinates of the mas 
ter station. Within the standard approach of GPS supported aerial 
triangulation these remaining systematic errors are introduced as 
additional unknowns and compensated in the bundle block adjust 
ment. Such an approach is not possible for the “simple” GPS/INS 
integration using a Kalman filter, as far as no informations from 
image space are used. In other words, every error that is not mod 
eled in the dynamic model of the filter will introduce errors in the 
georeferencing process. 
3 COMBINING GPS AND INS WITH AERIAL 
TRIANGULATION 
Similar to GPS supported aerial triangulation an integrated ap 
proach should be applied for the georeferencing of imagery by 
combining and utilizing as many informations from different sen 
sors as possible, i.e. GPS, INS, and informations from image 
space. This approach should • 
• enable a control of the georeferencing process by increasing 
the reliability of the whole system. 
• allow an operational processing in terms of 
- the number of required tie and control points, which 
should be less or equal compared to standard aerial 
triangulation with full frame imagery. 
- the potential of an automated processing. 
• enable a self-calibration of the camera. 
• provide a higher accuracy compared to direct georeferencing 
by GPS/INS integration, particularly if only data for the single 
image strips are available. 
3.1 GPS/INS data processing 
In contrary to the GPS/INS processing proposed i.e. by (Schwarz, 
1995), (Skaloud, 1995), (Sherzinger, 1997) within the algorithm 
presented here, no Kalman filter is used. Originally, this algorithm 
was designed for processing of the data from the DPA sensor sys 
tem - a three line push-broom scanner, that will be described in 
more detail in section 5 -, where inertial data are available only 
during the acquisition of image strips due to hardware restrictions. 
The lack of a continuous INS data trajectory prevents the standard 
Kalman approach starting with a static initial alignment for position 
and attitude. Therefore, the initial alignment has to be done in 
flight, during the motion of the aircraft. Usually, this in-flight align 
ment is obtained from gyrocompassing (mainly for roll and pitch) 
and the combination of GPS derived velocities to the inertial mea 
surements during aircraft maneuvers, which are performed to pro 
voke accelerations in all directions (mainly for heading). As there 
are almost no accelerations during the image strips, this method 
is not applicable to determine the initial attitudes, in especially the 
heading angle. 
Therefore the basic concept of the algorithm, which is presented in 
figure 1 is as follows. First a strap-down INS mechanization is per 
formed, which is supported by the GPS measurements. If there is 
no additional information available the initial offsets (accelerometer 
bias, gyro bias) of the inertial sensor are assumed to be zero for 
the first mechanization step of the INS data. The initial position and 
velocity are obtained from GPS. Assuming a normal flight, the ini 
tial orientation of the system will be close to zero for the roll u> and 
pitch angle ip. The initial heading k is obtained from GPS. Using 
the estimated initial alignment and the sensor offsets, the mech 
anization is done, whereas the INS derived positions are updated 
via GPS at every GPS measurement epoch. 
After integration the parameters of exterior orientation (position 
Xi,Yi,Zi, attitude u>i, ipi, «¿) are available for every measurement 
epoch i. The positioning accuracy is mainly dependent on the ac 
curacy of the GPS .positioning. The attitudes are mainly corrupted 
by a constant offset ojo,¥>o,«o due to the incorrect initial align 
ment. Additionally, there are some drift errors wi,(pi,Ki caused 
by remaining sensor offsets. These errors have to be determined 
and corrected (equation 1) to obtain corrected attitudes u>i,<pi,Ri 
and to get highest accuracies for the georeferencing. 
U>i + UJO + U)\ -t 
<Pi + <po + <Pi • t 
Ki + KQ + Kl ■ t 
(1) 
Equation 1 is a simplification of the true error behaviour. Additional 
errors introduced due to the correlations between the attitudes are 
not considered here. The effects caused by correlations are de 
scribed in section 4 in more detail. Nevertheless, applying this er 
ror model in an iterative process of a combined aerial triangulation, 
the best solution will be obtained after a few iteration steps. 
In addition to the INS error terms, the orientations are affected by 
the unknown misalignment Su, 5<p, 5k between the INS body b and 
the image coordinate frame p. 
3.2 Combined aerial triangulation 
The general idea is to perform an aerotriangulation of imagery in 
order to correct the position and attitude, which are provided from 
the GPS/INS module. Similar to the approach proposed by (Gib 
son, 1994), these terms contain INS error terms, as well as pa 
rameters for system calibration resulting from the physical offsets 
of the different sensors. Although the algorithmn was developed for 
the evaluation of line scanner imagery, the data of traditional frame 
sensors combined with a GPS/inertial module can be processed in 
the same way. 
Similar to the Kalman filter concept, the errors are grouped in an 
error state vector. This vector includes the navigation errors, the 
sensor noise terms and can be expanded by additional calibration 
terms. After mechanization the error terms are updated using the 
values estimated in the aerotriangulation step. Within this aerial 
triangulation the photogrammetric coplanarity (relative orientation) 
and collinearity (absolute orientation) are used for the estimation 
of the error terms. For reasons of simplification and flexibility the 
collinearity equation will be utilized in the following. 
4-5-3 
M I BU JB _ .„.jisjMMHMIMi 
1S: Sft