GPS - Position | 
update 
Y 
A | Gyro 
update 
y 
GPS - velocity 
update ] 
y 
Speed log 
update 
y 
| Radar contour 
  
  
  
  
  
  
Prediction with 
system model 
  
  
match gae 
  
mt target 
match = 
  
  
= scanner 
match update 
  
  
Integrated 
estimate 
e- = 
  
  
  
Figure 9: Flow of information within the Kalman filter 
imaging sensors, the extended Kalman filter also pro- 
cesses the measurements from a GPS receiver, a 
ultrasonic speed log and a directional gyro. 
Figure 9 shows the processing steps within one 
Kalman filter cycle. The filter uses a nonlinear system 
model with 9 states to perform the prediction of the 
ship’s state and its covariance. A detailed discussion 
of the system model can be found in [5]. The mea- 
surements of the different sensors are processed in 
separate update steps. This makes it easy to account 
for not available measurements by passing state and 
covariance unchanged through an update block. 
As GPS measurements, the position and velocity 
computed by the GPS receiver itself are presently 
used. They are transformed into global chart coordi- 
nates and used as separate updates. This is due to 
the experience, that the GPS velocity obtained from 
the receiver is more reliable than the position. GPS 
measurements are not continuously available to the 
navigation system. The reception of the satellite sig- 
nals is interrupted when the ship passes a bridge. 
Also satellites may be not visible when cruising in 
narrow valleys or near to trees on a river bank. Al- 
though the accuracy obtained solely from GPS in the 
standard positioning service is not sufficient for an 
72 
automatic cruise on inland waterways, GPS is an im- 
portant part in the integrated system. |t is used to 
stabilize the estimation of the position along the river. 
This is especially important for navigation in canals, 
when often only little information about the longitudi- 
nal position can be derived from image matching. The 
implementation of algorithms for processing GPS raw 
data and the application of DGPS is subject to current 
work within the project. 
The directional gyro measures the ship's heading 
relative to an unknown initial value. Alternatively a 
rate gyro measuring the yaw rate of the ship the may 
be used. The speed log measures the ship's speed 
over ground along the ship's heading. 
The matching updates are processed at the end of 
the update chain, because they need an initial po- 
sition and heading. By this order of updating, the 
measurements from GPS, gyroscope and speed log 
are already included in the initial position and head- 
ing for the matching processes. The initial values for 
matching with the radar image are extracted from the 
estimated state after the speed log update. The initial 
values for matching the laser scanner image is de- 
rived from the state after the radar updates. Within 
the matching updates the covariance matrix C, com- 
puted in the matching process, is used as the co- 
variance of the measurement noise. Thus the filter 
can account for the actual accuracies obtained from 
matching. 
The integrated estimation of the ship's state is given 
by the state vector after the laser scanner update. 
7 GENERALIZED STRUCTURE OF THE 
INTEGRATED NAVIGATION SYSTEM 
The generalized structure of the integrated naviga- 
tion system is displayed in figure 10. The integrated 
navigation system consists of a sensor processing 
level (dashed box in figure 10) and a control level for 
optimization and control. 
Generally the information obtained directly from the 
sensors is not sufficient for guiding the ship. As a con- 
sequence, a-priori knowledge of the ship’s dynamics 
and the features of the navigation environment must 
be contributed to the sensor information. The up- 
per part of the sensor processing level represents the 
real world, the ship, its navigation environment and 
all sensors available on the ship. The model world 
in the lower part of this diagram is a reproduction of 
reality in the computer by means of the available a- 
priori knowledge consisting of dynamic models for the 
own ship and other vessels and the electronic chart 
of the waterway. To match the real world with the 
model world the sensor signals are used. Difference 
signals are deduced by comparing the sensor sig- 
nals to equivalent signals generated from the model 
world. Examples for such difference signals are the 
corrections to the initial values of the image matching 
  
  
  
mm Um TT TTY