Figure 6: Example for matching stationary targets 
and chart 
tracking algorithm. Thus a correction of the position 
and heading of the own ship according to equation 
(2) is not needed. 
Based on the distance vectors a validation step is 
performed. All assignments with a distance larger 
than a certain threshold are assumed erroneous and 
the image measurements are discarded. Another 
condition can be stated to avoid wrong associations. 
A landmark of the chart can only be assigned to one 
stationary object. If there is more than one validated 
assignment, only the assignment with the smallest 
distance is taken into account for matching purposes. 
The distance vector is defined by the difference of po- 
sitions of two assigned points and therefore contains 
two dimensional displacement information. Thus the 
weighting of the measurements must be different from 
the algorithm described in section 4.1. In this case, 
the inverse of the position covariance matrix of the 
specific tracking filter can be used for weighting the 
measurements. With the computed distance vectors 
and the modified weighting matrices the corrections 
are computed as discussed in section 4.1. 
The result of matching stationary targets and elec- 
tronic chart is the correction As¥ and the covariance 
matrix Cs = [MT WM] ^! as a measure of the accu- 
racy obtained by the matching process. 
An example of this matching procedure is displayed 
in figure 6. The radar image was recorded at a trial 
70 
with the push tow L16 (185m, 10000 t) at the river 
Rhine near Nijmegen. The 'confirmed' radar targets 
are marked by rectangles. The size of the rectangles 
symbolizes the accuracy of the estimated position of 
the target. Their velocity is indicated by a vector start- 
ing in the middle of the rectangle. In the chart several 
groynes are visible extending from the river banks 
into the river. At the end of the groynes radar re- 
flectors are installed. These reflectors are stored as 
landmarks in the electronic chart. The radar reflec- 
tors are also tracked as indicated by the rectangles. 
Those targets marked with a cross are classified as 
stationary and therefore used for matching with the 
landmarks of the chart. 
4.3 Matching laser scanner image and electronic 
chart 
Within the integrated navigation system a laser scan- 
ner is used to sense the near surroundings of the 
ship. The scanner is mounted at the bow of the ship 
near the water surface and takes its measurements in 
one horizontal plane. The sensor covers an azimuth 
of 270 degrees and can detect objects in a distance 
of up to 50 m with an accuracy of 4 cm (1 e). The 
radar sensor has considerable disadvantages in de- 
tecting the near surroundings of the ship. The least 
detectable distance is about 15 m and the radial quan- 
tization of the digitized radar image is 3 m. Thus laser 
scanner measurements are essential for the naviga- 
tion in narrow canals, for entering locks or for docking 
manoevers. 
Although the coverage and the accuracy of laser 
scanner and radar are quite different, the images de- 
rived from both sensors are similar, because both 
sensors yield a map-like image of the surroundings. 
Thus the same matching techniques as explained for 
the radar in section 4.1 can be used for the laser 
scanner image. The implementation of the match- 
ing algorithm for laser scanner images is subject to 
current work within this project. 
Figure 7 shows an example of the laser scanner 
image. This picture was recorded when entering a 
lock chamber on the river Neckar. The black points 
are the laser scanner measurements, the grey line is 
the contour of the lock from the electronic chart. 
5 TRACKING OBJECTS IN IMAGE SEQUENCES 
Besides time invariant objects, e.g. river banks, the 
navigation environment scanned by the radar sensor 
also comprises foreign ships, anchored buoys and 
radar reflectors. In order to obtain information about 
the actual traffic situation, it is essential to reconstruct 
the trajectories of the latter objects, with respect to 
position, speed and orientation in relation to the own 
ship [7]. A bank of dynamic object models has to 
be processed in parallel, each model describing an 
   
  
Figur 
assumed 
not be ga 
sary to inl 
age sequ 
target tra 
Kalman fi 
tories giv 
sive Baye 
pret the : 
results in 
which is 
basic prol 
of hypoth 
sociations 
the implei 
Figure 
showing c 
Here wec 
respondir 
Kalman fi 
prediction 
tor. From 
the target 
of the nav 
Applyin 
one has 1 
ple from | 
bridges. 
in the ime