t) at the river 
’ radar targets 
the rectangles 
ted position of 
/ a vector start- 
2 chart several 
ie river banks 
/nes radar re- 
are stored as 
e radar reflec- 
he rectangles. 
2 classified as 
ching with the 
nd electronic 
1 a laser scan- 
ndings of the 
ow of the ship 
asurements in 
rs an azimuth 
in a distance 
m (1 o). The 
intages in de- 
ip. The least 
e radial quan- 
m. Thus laser 
or the naviga- 
or for docking 
jracy of laser 
le images de- 
because both 
surroundings. 
explained for 
for the laser 
of the match- 
is subject to 
aser scanner 
on entering a 
' black points 
ie grey line is 
lic chart. 
EQUENCES 
or banks, the 
radar sensor 
d buoys and 
mation about 
0 reconstruct 
th respect to 
jn to the own 
odels has to 
lescribing an 
  
  
  
  
  
1 1 i 1 L L 
  
Figure 7: Example of a laser scanner image 
assumed target track. This kind of information can 
not be gained from single radar images. It is neces- 
sary to interpret the chronological development in im- 
age sequences. This goal is achieved by a multiple- 
target tracking algorithm [1]. Methods applied are 
Kalman filtering [2] for the estimation of target trajec- 
tories given the data-association history, and recur- 
sive Bayesian tests [8] in order to sequentially inter- 
pret the arriving radar image measurements. This 
results in a set of contradicting hypotheses, each of 
which is associated with a certain probability. The 
basic problem originates from the exponential growth 
of hypotheses arising if all possible measurement as- 
sociations are considered. A detailed description of 
the implemented algorithms can be found in [6]. 
Figure 8 gives an example of a real radar image, 
showing one oncoming ship and several radar buoys. 
Here we can distinguish two gates for each track, cor- 
responding to the update and prediction steps of each 
Kalman filter. The line originating at the center of the 
prediction gate symbolizes the estimated speed vec- 
tor. From this image it can be seen, that by means of 
the target tracking technique, a reliable reconstruction 
of the navigation scenery can be obtained. 
Applying the tracking algorithm to inland shipping, 
one has to deal with radar echos arising for exam- 
ple from bushes or trees at the river banks or from 
bridges. Therefore a classification step is included 
in the image processing, where relevant objects for 
71 
  
  
  
  
Figure 8: Example for multiple-target tracking meth- 
ods 
tracking are selected. Within this step, the informa- 
tion of the electronic chart is used. The distances 
from a radar object to the nearest objects in the chart 
provide additional classification information. So it is 
possible to decide, whether a radar object is outside 
the river and thus irrelevant for tracking. The detec- 
tion of radar buoys and landmarks and the treatment 
of echos arising from bridges or overhead lines can 
also be improved by using the chart information. Thus 
the application of chart information improves the effi- 
ciency of the tracking algorithm. 
6 INTEGRATION OF MEASUREMENTS 
The integration of results obtained from different 
matching algorithms and the measurements of other 
sensors is performed by an extended Kalman filter 
[2]. Within the filter, the different measurements are 
weighted according to their actual accuracy and an 
integrated estimate of the ship's state is computed. 
Designing the filter, one has to take into account that 
the accuracy of the measurements may vary with 
time. Measurements may even be not available tem- 
porarily. When for example no landmarks are visible, 
the matching of stationary targets and landmarks 
cannot be performed. Also some sensors may not 
be installed on a specific ship. 
In addition to the measurements derived from the