2-2-3 
tW0! ^systems 
im c °ntains an 
^isation, 
yros and thr% 
s aut °nomousi y 
^ KiSS. ^ 
laced in the left 
ial ^nsors are 
iati °ns, aiding 
' ^ese are zero 
cities from an 
iti°ns whenever 
:s using pseudo- 
hded with 5 Hz. 
iie the coverage 
v 20 per cent. 
rack of the van 
îochrome CCD- 
he situation in 
ing system with 
n a roof rack of 
0 50 min front 
both directions, 
1 by the system, 
ite and on the 
IS video system 
K synchronised 
n. 
3 POSTPROCESSING CONCEPT 
i The evaluation of all sensor data is carried out after the data 
i acquisition. The basic structure of this post-processing is outlined in 
; figure (3). It consists of two independent parallel computations: the 
I! determination of the trajectory and the object recognition. 
I| Subsequent to this step is the setup of the data base, the 
! determination of the alignment elements and the object positioning. 
Figure 3: Overview of the post-processing concept 
The determination of the trajectory is carried out by a cascaded 
filter system. The structure of this system is outlined in figure (4). It 
consists of two pre-processing steps and the main filter. 
! Figure 4: cascaded filter system with two pre-processing steps 
and central Kalman-Filter (including sensor signals) 
3.1 Pre -filtering of the trajectory observations 
In the first step of pre-processing the original measurement values 
are corrected and converted to pseudo-observations, like air 
pressure differences from the barometer readings to. height 
differences. 
In this step of the post-mission data processing it is also necessary 
to transform all the data from the individual sensor coordinate 
systems into a common reference frame. This is defined as the body 
system of the van with its origin given by the reference point of the 
IMU and axes parallel to the main axes of the van. This 
transformation is applied for both, position- and velocity 
determination. 
The second pre-processing step consists of smoothing all pseudo 
observations. The GPS Kalman filter is a completely independent 
Kalman filter, where the double-differenced pseudo-ranges from 
code measurement and the triple-differenced carrier phase 
measurements (phase rates) are processed as observations. Since 
changes in position can considerably be smoothed by GPS Kalman 
filtering all other pseudo-observations have to undergo filtering and 
smoothing algorithms in order to obtain the homogeneity of input 
data for central Kalman filtering. The filter-smoother automatically 
locates outliers and provides estimates of the precision. In order to 
obtain an optimal control on sensor signals and pseudo 
observations, respectively, all sensors are pre-processed separately 
and subsequently combined in a central Kalman-Filter with feed 
back. 
3.2 Central Kalman Filter 
The three-dimensional position as well as the velocity from GPS, 
the three rotation angles of the Inertial System, the velocity in 
direction of motion from the odometer and last not least the height 
change from barometer readings are introduced into the Kalman- 
Filter as observations. The observations are summarised in table 1. 
observations 
x, y, h 
3 D position from GPS 
vx,vy,vh 
3 D velocity from GPS 
<px, <py, <pz 
Azimuth from INS 
v odo 
velocity from odometer 
dh 
height differences from barometer 
Table 1 : The observations of the main Kalman Filter 
The theoretical and mathematical structure of the Kalman-Filter is 
an expansion of the one comprehensively described in (Wang 1997) 
and (Sternberg 1996, Sternberg 1998a) and therefore we have not to 
go into further details in this article. In contrast to conventional 
navigation filters being designed for the estimation of the error - 
state of the system, the position is directly estimated in this filter. 
Furthermore the azimuth, the tangential velocity of the system 
(equivalent to the velocity in moving direction) and the normal 
acceleration, which is perpendicular to the moving direction, are 
estimated in the main filter.