”The integration of inertial and GPS satellite techniques 
currently offers the best potential for implementing georef- 
erencing systems...” (Schwarz, 1995). GPS has long term 
stability with an accuracy of one or two decimeters if Differ- 
ential GPS (D-GPS) is applied. The INS has a high relative 
accuracy for the derived position and attitudes but because 
of the integration process drift can not be avoided. 
The design of a Kalman filter is considered to be the most 
promising technique for GPS-INS integration. Because in 
our case existing tools for an independent processing of 
D-GPS and INS are used we decided to choose a more 
pragmatic solution to exploit D-GPS for eliminating drift in 
INS positioning. This is done by the following steps: 
1. Estimate approximate INS drift parameters by using 
polynomials. 
For modelling the attitude drifts a first order polyno- 
mial was sufficient. But for the position a fourth order 
polynomial has given the best results. This indicates 
that the initial alignment of the INS body frame with 
the computational frame was not satisfactory. 
2. Determine the datum transformation between the 
GPS and the INS coordinate frame and adjust the 
high frequency (235 Hz) INS data to the low fre- 
quency (1 Hz) GPS data. 
The datum transform is solved by 7 parameter trans- 
formation. After that GPS and the residuals between 
GPS and INS are given in the local INS frame. This 
residuals are used for a refined adjustment of the INS 
data to the GPS trajectory. 
The result of this processing is illustrated in figure 4. The 
plot shows the resulting height component for a part of one 
strip. 
GPS-heights [m] for each second (1 Hz) 
  
2343.0 
  
  
  
Figure 4: Height-component of a part of the flight path 
4.2 Image Rectification 
The recorded images displayed by small parts in figures 5 
and 7 show distortions which are mainly caused by attitude 
variations. Correcting for this movements is done by using 
the orientation of each line derived from the GPS-INS data 
as described above. 
Figure. 5: 
   
Original image Figure 6: Rectified image 
(1024? Pixel) 
  
144 
Figure 8: Zoom of rectified 
nal image (upper row 144%, image 
lower row 96? Pixel) 
Figure 7: Zoom of origi- 
A rectification plane perpendicular to the height axis of the 
local coordinate system is defined. This horizontal plane 
is located in a mean flying height above ground (Z-plane). 
Then “approximate vertical” images are produced based on 
the direct method for image rectification as indicated in fig- 
ure 9. Each pixel in the image space is projected on the 
Z-plane by using the orientation parameters. The resulting 
points are irregularly distributed in this plane and carry the 
grey value of the corresponding image points. The transfer 
to the regular grid is solved by interpolation using weighted 
averaging of grey values of all neighbouring points within 
a certain radius. The weight is chosen reciprocal to the 
distance between the grid point and its neighbours. If no 
neighbours are found around a grid point, as this is the case 
outside the borders of the image, a background value is as- 
signed to this pixel. Examples of the rectification are shown 
in figures 6 and 8. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
4.3 
Our 
tispe 
field 
extr: 
builc 
of d 
are i 
beei 
and 
worl 
pair. 
of ol 
With 
tispe 
asse 
(19€ 
mad 
For 
whic 
imac 
in fic 
For ! 
isa 
dure 
men 
mov 
the « 
ing « 
men 
anu 
ple, 
The 
geth 
11. 
tatio