399.300
Figure 2: Example of an electronic chart
dian spirals. This type of interpolation is especially
appropriate for a river and results in a small number
of interpolation points.
The appearance of objects represented in the chart
usually depends on the water level. This dependance
is linearly interpolated in the chart by assigning a
slope of the embankment to every spiral or polygon
element. The value assigned can also be interpreted
as a measure of accuracy of the element for image
matching purposes.
An example of a hardcopy of the electronic chart
display is shown in figure 2. Two harbors and a bridge
can be easily distinguished. In the river, the river axis
and the limits of the navigable water are plotted as
solid respectively dashed-dotted lines.
More details about the implementation of the chart
are given in [4].
4 MATCHING IMAGE DATA AND ELECTRONIC
CHART
Within this section, a least squares technique for
matching images and the electronic chart is ex-
plained. Input to the matching algorithm are predicted
values of the ship's position and heading. Corrections
to these predicted values are computed by minimiz-
ing a weighted sum of distances between image and
electronic chart. These corrections and their accu-
racies are used as measurements in a Kalman filter
described in section 6.
The matching technique is used for matching radar
as well as laser scanner images with the electronic
chart. Furthermore, stationary targets detected by a
multiple-target tracking algorithm can be matched to
landmarks in the electronic chart by the same tech-
nique.
4.1 Matching radar echo contours and elec-
tronic chart
The process of matching the radar contours and the
electronic chart starts with an estimate of position and
heading based on previous matching steps. For the
current radar image, a prediction of the ship's position
and heading has been computed according to the
mathematical model of the own ship. Also actual
measurements from other sensors like a gyroscope
or GPS may have been processed by a Kalman filter
algorithm, so this information is already integrated
in the initial values of the matching process. The
initial values consist of the global coordinates s,(.
and s,(_) of the position of the imaging sensor and
the global heading v... of the ship. The acquisition of
one radar image lasts some 2.3 seconds, the time for
one complete turn of the radar antenna. As the ship is
in general not stationary, the coordinates obtained by
the matching process are assigned to the time when
the antenna beam is pointing in the ship's heading
direction. As mentioned before, the purpose of the
matching process is the computation of corrections
As}
As = As; (1)
Av*
to the initial position and heading.
The standard radars used on commercial vessels
generate a map-like image of the surroundings. The
antenna beam has a horizontal beam width of 0.5 to
1.0 degrees, the vertical beam width is about 15 to 20
degrees. The antenna performs one complete turn
within 2.3 seconds. During the turn, radar pulses of
50 ns length are sent out, the pulses are repeated
with a maximum frequency of 3 kHz. Thus the raw
radar image consists of about 7000 radial beams.
The echo signals received by the antenna are digi-
tized, filtered and transferred into the main memory
of the computer. When filtering the raw image, a re-
duction of the number of rays is also performed. The
radar data available in the computer are made up as
a binary image consisting of 720 rays with 256 radial
pixels. The radial quantization can be chosen as 3 or
6 m, thus resulting in a maximum sensing distance of
768 or 1536 m. Currently the interface to the radar
is redesigned to enable a maximum number of 4096
radial pixels.
On the digi
cessing steps
segmented in
uous echo ar
area of the o
determined.
Relevant €
gained from
ate radar obj:
these radar c
cated closely
Thus a furthe
in about 200
Figure 3 sho
before match!
given in a pc
ship’s body, tl
points within
3 the nearest
chart with its
The distance
contour. Ba:
contour a vali
exceeds a ce
and chart is
ment is discal
from noise in
ously not con
validation ste
available for 1
of the neares
taken into a
and heading
expressed by
ST,
Sh; =
Vi
The correc
interpolating .
of the last ra
matching pro
suring time o
Based on
derived:
| —riSin(
rr; Cos
a k
= k
As, Ash
the initial pos
equations w
into account
der terms, e
66
EEE mari