2-2-4 
The mathematical model for the two-dimensional case was pointed 
All observations are processed in a forward and a backward filter. 
The estimated state-vectors from both filters are combined in an 
optimal smoothing algorithm. Besides the digitised trajectory in 
terms of coordinates, attitude angles and velocities at every PPS- 
second, which coincides with the moment of shooting of the CCD- 
cameras, the expanded filter provides estimates of calibration 
parameters for the sensors. The accuracy of the filtered points is 
about 0.5 m in a UTM coordinate system based on WGS 84. In the 
final step the sequence of points is replaced by a smooth line 
consisting of geometrical elements (straight sections, transitional 
and circular curves) forming the alignment of the road. 
time in seconds 
Figure 6: Results of the filter compared with OTF positions 
From the direction of movement at the outage the difference of 0.6 
m in the east component can be attached to a scalefactor of the 
odometer. The length of the outage is about 1200 m, which results 
in a scalefactor of about 0.05 %. The difference in the north 
direction, which is perpendicular to the moving direction, 
correspond to a gyro drift of about 0.08 ° / h. Both values are in the 
error budget of the sensors. 
With the smoothing algorithm the divergence can be reduced below 
3.3 Filter results 
0.4 m. 
To analyse the filter behaviour and the filter results control points 
were established on the former airport of Neubiberg, near Munich. 
This airport is suitable for the test, because it features optimal 
conditions for GPS measurements and optimal geometry for the 
vehicle path like long straight lines with no height differences. The 
airport and the path of a driven test route is shown in figure 5. 
Figure 5: Airport Neubiberg with vehicle path 
In this test at least 5 satellites were in view, enabling the 
computation of OTF solutions as a reference. A total outage of all 
satellites for 60 seconds was simulated to check the main filter. 
During the outage the vehicle is on the taxiway, which is oriented in 
west-east direction. In figure 6 the position differences between the 
OTF positions and the results of the forward filter, of the backward 
filter and of the smoothing algorithm respectively are shown. 
3.4 Object measurement 
By screening the video, the objects of interest are identified and 
flagged for measurement. Since the video is synchronised with the 
CCD-cameras, it is easy to find the flagged objects in the digital 
images. A photogrammetric software has been developed which 
supports the measurement of the location of objects in all images, 
where they appear, and performs a robust adjustment to determine 
the 3d-coordinates. The self calibration of the pair of cameras 
provides the parameters of the inner and the relative orientation. 
The transformation of the coordinates from the camera system into 
the body system requires rotations and translation using parameters 
which have been resolved by a test field calibration. The final 
transformation into the world coordinate frame is based on the 
position and the attitude of the system at the moment of exposure, a 
result of the Kalman-Filter. The standard deviation of the relative 
position of the objects is smaller than 5 cm as shown by Klemm 
(Klemm 1997).The accuracy of the absolute position including the 
computed trajectory is about 0.5 m. 
The graphical user interface for the photogrammetric measuring 
procedure is shown in figure 7. An interface to a geographic 
information system is available to transfer the positions and to add 
the attributes required for further use. 
'9 
** 
Figure’ 
4 
In a pilol 
demonstartf 
of the highv 
lights, bicy 
districts. Th 
For this aii 
directions.' 
measuremei 
total and 
transformat: 
coordinate 
hour of dri 
forests wen 
The object 
the environ 
The accura 
different n 
position W; 
sensor cal: 
to CCD si 
well as the