International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 1
Number 2
June 2007
153
Intelligent VTS
W. Filipowicz
Gdynia Maritime University, Poland
ABSTRACT: In this paper the author depicts new approach involving fuzziness in navigational situation
assessment. Nowadays operator at maritime traffic monitoring station is assumed to have access to a great
amount of data. The data comes from different places and multiple of sensors. Properly associated data enable
the operator to approximate congestion for each restricted and considered as vital regions. Ship’s presence
function within a confined area defines a non-empty bounded closed interval. It can be denoted by the earliest
and latest bounds of the closed time interval at a given possibility level. To assess situation within any
confined region one should calculate maximal sum of safety factors present within forecast imprecise slots of
time. Safety factors themselves are also fuzzy, imprecise values.
1 INTRODUCTION
Vessels traffic is monitored using different appli-
ances. Main monitoring aim is to check that
everyone obeys imposed rules and traffic separation
is not violated. The measure introduced within
Vessels Traffic Systems (VTS), mostly used radar
stations, significantly contributed to reduction of risk
of collision and improved environment safety
standards. Nowadays radar surveillance can be
supplemented by other sources of data. AIS
(Automated Identification System) which is about to
be introduced seems rather limited in its ability. The
throughput as well as an overall functionality seems
do not meet expectations related to this new
technology. For this reason there are attempts to
create ad hoc networks to transfer even more data
among ships using wireless transceivers. Dynamic
schemes could include shore hot spot stations to
open access to the internet and transmit data to all
interested parties.
Multiple sources of data create new challenge
regarding data association. The challenge is met by
appropriate but still emerging technology called data
fusion. By means of fusion, different sources of
information are combined to improve the perfor-
mances of a whole system. The most obvious
illustration of fusion is the use of various sensors to
detect objects. Available data could be subject to
various processing and the results of different
procedures may be further combined. Objectives
such as detection, identification and tracking can be
achieved thanks to data fusion. Data fusion produces
high quality, enriched and reliable sets of data. Such
sets are necessary for further processing to create
foundations for decision making process.
Data fusion is a process dealing with the
association, correlation, and combination of data and
information from multiple sources to achieve refined
position and identity estimates for observed entities,
and to achieve complete and timely assessments of
situations and threats, and their significance. These
can contribute to further improvement of the safety
standards in particularly within restricted waters. It is
also assumed that VTS operator is able to have
access to these reliable and enriched data, which
create foundation for implementation and execution
of a policy aimed at prediction and traffic control
within the region.
Avoiding local congestion one can reduce number
of encounters and furthermore potential risk of
154
collision (Filipowicz et al. 2005). To introduce such
measures a few assumptions are to be made. First
there must be all data fused with unlimited data flow
within wide area sea routes schemes. All local
branches of VTSs are to be networked and all
available data regarding traffic along with local
conditions easily exchanged. Second there is a
decision-making body within VTS structures.
Databases are to be implemented and decision
problems to be formulated and solved. The last
comes along with proposal of the set of assessment
criteria and delivering necessary tools to decision
maker.
In order to take adequate decision one has to
compare a handful of parameters of different types.
Basically there are crisp and fuzzy values to be taken
into account. For particular vessel and each route,
she is assumed to take scheduled traffic is an
important factor. For VTS operator traffic
encountered within each restricted and considered as
crucial regions can usually be foreseen. Quality of
the forecast depends on data precision.
To process imprecise inputs one has to introduce
interval arithmetic with possibility level selected.
Ship’s presence within confined area defines a
bounded closed interval of time. It can be denoted by
the earliest and latest limits of a time slot with a
given possibility level. To assess navigational
situation within any confined region one should
calculate maximal sum of safety factors present
within at any moment. Safety factors, which reflect
tonnage of a ship as well as carried cargo, are also
likely to be fuzzy values. Subjective assessment of a
term like “large vessel” should be perceived rather as
a range of values than a single, crisp one.
Data fusion technology will be discussed in the
next chapter. Fuzzy or interval arithmetic will be
shortly presented in the further chapter. In the second
part of the paper fuzziness in the vessels traffic
engineering will be presented.
2 DATA FUSION
Modern VTS logical structure should integrate of
traditional and advanced surveillance appliances,
communications means, computer, and other techno-
logies for purposes of improving navigation safety
standards at waterways. The overall integration
techniques are embraced within data fusion. Data
fusion is a scheme of collecting, processing and
enrichment of informational aspects of available
data. Multiple level model (figure 1) extends from
raw sensor data up to situation refinement and final
decision making.
Level 1: Raw data sources
(radars, AIS, ad hoc networks).
Signals refinement
Level 2: Objects refinement
Level 3: Situation refinement
Level 4: Decision making
Fig. 1. General scheme of data fusion in maritime traffic
engineering
Multi-source data fusion enables deliver
information that is characterized by (Sarma et al.
1991):
− increased confidence - more than one sensor can
detect the same object
− reduced ambiguity - joint information from
multiple sensors reduces the set of hypotheses
about the target
− improved detection - integration of multiple
measurements of the same target increases
possibility of detection
− extended coverage and reliability - one sensor can
work when or where another sensor is out of
order or remains beyond the range
Data fusion comprises four levels. Signal, object
and situation refinement are carried out at the lower
levels. Situation recognition and identification
deliver reliable and adequate set of information for
the highest level where decisions are made and
recommendations issued. Table 1 shows main fusion
levels, their methods and techniques.
Table 1. Main fusion levels, their methods and techniques.
_________________________________________________
Fusion level Methods and main tasks Techniques
_________________________________________________
Level one signal refinement, Kalman filtering
position estimation
Level two object refinement, Bayesian methods,
pattern recognition Dempster-Shafer
reasoning
Level three situation refinement fuzzy logic,
modeling
Level four artificial intelligence, MADM Decision
optimization, Making
best passage (TOPSIS, AHP)
recommendation
________________________________________________
2.1 Level One Fusion
Association of multi-sensor raw data is the main task
for the first level of data fusion. It is supposed to
correlate sets of sensor outputs. As a result the data
can be used to estimate a target’s position, course
155
and velocity. In order to remove noise from sensor
signals many systems employ Kalman filtering
technique.
Kalman filtering produces data that estimate the
targets positions. Consequently smoothed positions
coordinates values enable better estimates of velocity
and course (Linn et al. 1991). The accuracy, the
tolerance of each sensor’s output accuracy can be
estimated and assigned. In such case object’s
approximate position can be roughly forecast.
Kalman filtering delivers ability to define limits
within which an object will be located.
2.2 Level Two Fusion
It is said that at level one sensor signals are refined.
Level two data fusion is considered to be an object
refinement level. It delivers more processed and
meaningful data. To create a model of uncertain
system states by consolidating and interpreting
overlapping data delivered by different sensors
Bayesian decision theory is widely used. Dempster-
Shafer Evidential Reasoning (DSER) is an
alternative to Bayesian approach. It is known as a
generalization of Bayesian method that offers a way
to combine uncertain information from various and
unreliable sources. It works with fuzzy data,
perceived as intervals of confidence instead of
unique probability value.
Neural network technology can be also used at the
second level of data fusion. Neural networks
produce results that incorporate input from various
information sources. A neural network consists of
processing nodes that collect and compare data.
Nodes also called neurons are interconnected and
weights and applied to their outputs, which are then
forwarded to the successor nodes. Usually a neuron
has many inputs, but it has only a single output. It
has attributed equations that define what the output
would be like for given inputs. For neural network
there must be learning scheme. During the learning
period weights are adjusted according to the
stimulation the neurons receive. During the learning
process each neuron must be taught establishing the
proper relation between their inputs and generated
output. The network is taught through the
observations of discrepancies between expected and
achieved results which cause modified weights
assigned to each neuron until an expected behaviour
is obtained.
2.3 Level Three Fusion
The third fusion level is supposed to refine situations
which result from detected objects movement. In the
discussed navigation field it should create picture of
what will take place within particular areas. In
restricted waters with heavy traffic it is important to
avoid local congestions in routes crossing regions.
It is quite often that values cannot be categorized
using strict range limits. For example it is not
practical to expect that a vessel will be crossing an
area starting at time t
1
and ending at t
2
. Instead one
must specify that she will arrive sometime around t
1
and depart from the area at about t
2
. Fuzzy logic is a
type of theory that mathematically describes
imprecision and is widely used at data fusion process.
This level also employs artificial intelligence
methods. Computer Expert Systems emulate the
behavior of a human expert. They consist of two
components: inference engine and the knowledge
base. The inference engine performs search through
available possibilities in order to arrive at
appropriate conclusions. The knowledge base is the
set of facts and rules. These rules are usually in the
form of “IF-THEN” statements. Modern expert
systems are able to cooperate with knowledge bases
using fuzzy logic.
2.4 Level Four Fusion
Level four should be discussed taking into account
specificity of a particular field of interest. There are
a few problems, which still remain unsolved, in
vessels traffic engineering. First is a VTS supervisor
problem when he is asked for advice on best
possible passage for particular vessel. The problem
is like “I am a VLCC scheduled to reach reporting
point at some time. Please advise me the best route
or an option for the passage. Should I delay in order
to pass disturbed as little as possible”. The question
is probably addressed to the VTS operator of the
local control station. At the other side the advisory
body of a VTS is supposed to be interested in such
sporadic requests, but it also should be engaged in
everlasting job of traffic allocation in order to avoid
local congestions.
3 FUZZINESS
Often it is desirable to process imprecise or
approximate values. Fuzzy numbers are useful when
dealing with imprecision (Kaufman 1991). Fuzzy
numbers are sets of the real figures that are treated as
intervals. Their geometrical images are of triangular
or trapezoid shapes and are called as membership
functions. Triangular fuzzy value is referred as to
triple of figures (a, b, c), trapezoid one as to quad (a,
b, c, d). Fuzzy values express possibility of being
within given range of values. They start from zero
(lack of possibility) and reach maximum possibility
156
level equal to one. Possibility is different than
probability. Cumulated, integrated probability
distribution must produce one as a final result. The
requirement is not valid when dealing with
membership functions.
Arithmetic of fuzzy values is related to the
possibility level (α) and is based on α-cuts. The
α-cuts of fuzzy numbers represent possibility levels
and are closed intervals of real numbers.
Mathematical operators on fuzzy values are applied
to the boundary values of α-cuts. The idea of
exploiting this fact delivers straightforward
analytical method of dealing with non-linearity as a
result of multiplication of a fuzzy values (Filipowicz
2006).
3.1 Fuzzy Safety Factors
Traffic should be classified taking into account gross
tonnage of a vessel and a kind of cargo she has on
board. Safety factors have been introduced to enable
classification of vessels. In general approach
environmentally dangerous freight and huge tonnage
increase the factor. As it was proposed the factor
vary on an integer scale such that the higher the
number the more serious the consequences of an
accident. Small value was assigned to small craft
without dangerous cargo. The largest value was
reserved for huge crude carriers. It was assumed that
safety factor is easily assigned to every ship and
classification is free from any ambiguity. Since small
and huge are imprecise linguistic terms they should
be treated as fuzzy values. Suggested assignment of
imprecise, fuzzy safety factors to selected classes of
crafts is presented in table 2.
Table 2. Fuzzy safety factors assignment.
__________________________________________________
Cargo Tonnage of craft
Small Medium Large Very Large
__________________________________________________
ND SF (0, 0, 1) (0, 1, 2) (2, 3, 4) (4, 5, 6)
Abr. S M L VL
k 1 2 3 5
MD SF (3, 4, 5) (5, 6, 7) (6, 7, 8) (8, 9, 10)
Abr. S&MD M&MD L&MD VL&MD
k 4 6 7 9
D SF (7, 8, 9) (9, 10, 11) (10, 11, 12) (12, 13, 14)
Abr. S&D M&D) L&D VL&D
k 8) 10 11 13
VD SF (11,12,13) (13,14,15) (14, 15, 16) (15, 16, 16)
Abr. S&VD M&VD L&VD VL&VD
k 12 14 15 16
__________________________________________________
General scheme of assignment is based on four
classes of ship’s tonnage: small, medium, large and
very large. There are four categories of cargo:
normal (ND - no dangerous), mildly dangerous
(MD), dangerous (D) and very hazardous (VD).
Table 2 contains proposal of assignment. Safety
factors (SF) should be divided by 16 for the sake of
normalization. Abbreviations used in examples
included in the paper as well as k value to be applied
with formula 1 are also presented in the table 1.
Final assignment embraces distortion caused by
supremacy of tonnage over cargo for the adjacent
groups of carried load. Very large vessel without
hazardous cargo has greater factor then small ship
with mildly dangerous material on board.
For given k, indicating number included in table 2
normalized fuzzy safety factor can be calculated
using formula (1).
(1)
3.2 Fuzziness in maritime traffic engineering
Whenever restricted area passage is considered
traffic encountered at routes crossings is to be taken
into account as important factor. Awareness of other
ships significantly increases wherever collision
avoidance is hampered. To assess navigational
situation within confined areas approximations
regarding all scheduled traffic are to be taken into
account.
Due to unforeseen deviations from intended track,
bad estimation of main engine performance and
collision manoeuvres seafarers always use estimated
time of arrival. For the same reasons ship’s presence
within any area should be treated as trapezoid fuzzy
value. The values consist of estimated earliest and
latest time of arrival as well as earliest and latest
possible time of departure from the region. Situation
within confined area are vital from safety point of
view. Figure 2 presents example with a few crafts
that are scheduled to pass restricted area. There are
four vessels that are very likely to encounter within
the region where any collision avoidance manoeuvre
is seriously hampered. Vessels types were classified
as: S&D, L&D, S&MD, S&D (see table 2). Intended
courses of the vessels are shown at figure 2. Sea and
weather condition along with tonnage and speed of
each craft are given and subsequently fuzzy
timetable of crossing the area were estimated,
example results are presented at figure 3.
As it was already mentioned time frame of ship
crossing an area can be defined by trapezoid fuzzy
value. The membership function starts at the earliest
157
time of arrival and ascends to the appropriate latest
moment. This part of the function represents
entering phase, its inclination depends on the initial
distance from the area, weather condition, tonnage as
well as on propulsion ability of a particular craft.
Collision avoidance manoeuvres (if any) introduce
further delays. Right hand side of the function
represents departure phase and consists of a leg
joining earliest and latest possible time of leaving
the region. Figure 3 shows example membership
functions for situation presented in figure 2.
S
1
S
2
S
3
S
4
Fig. 2. Vessels that are likely to encounter within confined area
create potentially dangerous situation
Let us consider situation presented at figure 2.
We assume that one of the vessels marked as S
2
seeks for advice on best passage option (route to be
taken and/or time frame suggested). In the presented
situation, according to the COLREGS regulations
(see International Maritime Organization website
www.imo.org/Conventions/ for details), she is
supposed to be give-way with respect to S
3
and S
4
.
Her status referring to S
1
is stand-on. The status of
the vessel of interest with respect to each another
yields fuzzy weight factor. Table 3 embraces all
close approach situations and suggested normalized
fuzzy weight factors.
The most uncomfortable is crossing encounter
with give-way (as stipulated by COLREGS) status.
The potential of the situation gets even worse where
there is confined room to carry out collision
avoiding manoeuvre. For this reason respective
weight coefficient is the highest one.
Table 3. Close approaches and their fuzzy weight factors.
________________________________________________
Vessel status and its
Encounter type abbreviation Fuzzy weight
________________________________________________
Crossing Give-way, CGW (0.8, 1, 1)
Crossing Stand-on, CSO (0.4, 0.6, 0.8)
Overtaking Give-way, OGW (0.1, 0.3, 0.5)
Overtaking Stand-on, OSO (0, 0.1, 0.2)
Head-ons Give-way (0, 0, 0.1)
(each vessel), HO
_______________________________________________
Abbreviations used at figure 3 stand for:
A
α
l Si
- earliest time of arrival of the ship S
i
to
the given area taking into account α
possibility level
A
α
u Si
- latest time of departure of the ship S
i
from the given area taking into account
α possibility level
f
L
Si
(t), f
R
Si
(t) - left and right hand side boundary of the
presence function for ship S
i
(linearity
assumed)
f
Si
(t) - overall presence function for ship S
i
within given area
t
m
- time for which maximum fuzzy
congestion is found
ship
f
S3
(t)
time
S1
S2
S3
f
S1
(t)
α
f
S2
(t)
α
1Su
A
α
1Sl
A
f
S2
L
(t)
f
S2
R
(t)
t
m
S4
f
S4
(t)
Fig. 3. Ship’s presence within restricted area can be perceived
as trapezoidal fuzzy values. To assess passage condition one
has to scan entry phase and staying within time slot
Table 4. Navigational condition assessment for example area.
__________________________________________________
Ship Fuzzy SF f
Si
(t
m
) Status/
α
-cuts
Weight of the product
__________________________________________________
S
1
(S&D) (7, 8, 9) 1 CSO/ [0.145, 0.436]
(0.4, 0.6, 0.8) [0.192, 0.367]
[0.244, 0.303]
0.273
S
2
(L&D) (10, 11, 12) 1 own ship/ [0.727, 0.909]
(1, 1, 1) [0.764, 0.873]
[0.800,
0.836]
0.818
S
3
(S&MD) (3, 4, 5) 1 CGW/ [0.073, 0.273]
(0.8, 1, 1) [0.112, 0.236]
[0.157, 0.200]
0.182
S
4
(S&D) (7, 8, 9) 0.8 CSO/ [0.223, 0.436]
(0.8, 1, 1) [0.281, 0.407]
[0.335, 0.378]
0.363
__________________________________________________
[
1.178, 2.055]
TOTAL [1.349, 1.883]
[1.536, 1.717]
1.636
__________________________________________________
158
To assess passage condition for given vessel one
has to scan at least her entry phase and „staying
within” time slot. The slot for the situation presented
in figure 2 is marked in figure 3 with rectangular
shape. Numerical calculation for the situation is
included in table 4. Column „
α
-cut of the product”
in this table contains boundary values for possibility
levels α respectively equal to 0, 0.4, 0.8 and 1.
Final result shows total non normalized and non
regular fuzzy value. To compare such values one has
to normalize and defuzzify them. Defuzzification
converts imprecise intervals into crisp value. Many
fuzzy number ranking methods can be used.
However, no one can rank fuzzy numbers
satisfactorily in all cases and situations.
3.3 Membership functions estimation
To foresee encounter numbers a timetable of arrival
at given points are to be constructed for each
scheduled vessel. Timetable of passage, for each
vessel, and for given area is a vector of fuzzy slots,
which are quads of values that define membership or
“presence in the region” function. Earliest arrival
time (AlE) and the latest departure time from the
area (AuL) of the particular vessel are reference
values that create a time frame. The frame is to be
scanned to evaluate crossing condition.
Shapes of the presence functions, associated with
difference between earliest and latest moments of
arrival or departure primarily depend on necessary
deviation from the prescribed trajectory. To foresee
what will take place within given area one has to
construct (or learn) all presence functions.
Approximate numbers of collision avoidance
manoeuvres are to be counted since they influence
inclinations of ascending and descending slopes of
the functions. These can be estimated based on
simulations. Modeling and simulation computer
environment is necessary for implementation of the
discussed idea.
Basic assumptions of the environment concept
embrace:
− there is a module with interface enabling defini-
tion of the routes scheme (arrival and turning
areas). Route consists of legs linking turning
areas
− there is an interface enabling input of initial
positions and intended route for all crafts
− there is an interface enabling ship domain(s)
definition and selection, there must be database of
domains available
− decision regarding collision avoidance manoeuvre
is based on domain penetration by another vessel.
Adequate manoeuvre, as stipulated by
COLREGS, is carried out if required. There must
be an option of passing through without looking
at others (no collision avoidance manoeuvres
carried out)
− Close quarter approaches are classified and
recorded, all data necessary for further analysis of
ships involved in close approach are also stored
− movement along prescribed trajectory is double
screened random Markovian process.
For the sake of membership function estimation
all initial positions of all scheduled traffic for given
moment, using all available sources and techniques
must be calculated. Schedule traffic destinations and
intended or assumed routes are to be fixed. Average
values of their engines performances must be
gathered.
Two steps of analyses „for presence within”
functions estimation are suggested. At first earliest
arrival times are calculated and close quarter
situations registered. Shortest possible paths
assuming no violation of the separation schemes are
taken into account. During simulation with collision
avoidance option switched off encounters are
detected when safe distance limit is violated. All
close approaches are recorded for further analysis.
Data of ships involved in close approach are also
stored. Two ships are registered being involved in
close approach when it first occurs, their subsequent
mutual positions are not considered unless category
of encounter is changed. Categories list of encoun-
ters embrace: meeting, overtaking and crossing,
which is further subdivided regarding angle of
crossing.
At the next step latest arrival times are estimated.
Selected domain and all encounters are taken into
account and numbers of collision avoidance mano-
euvres are estimated for each of the vessels. It is
assumed that necessary collision avoidance is carried
out whenever adopted domain is penetrated by
another vessel. Such manoeuvre influences presence
function within all regions remained for passing
along given route. All these lead to estimation of
latest moments of presence functions.
4 FINAL REMARKS
Data fusion approach to deal with multiple data
sources in vessels traffic engineering was briefly
presented. Navigational situation within restricted
regions were characterized using fuzziness. Ships
fuzzy safety factors related to gross tonnage and sort
of carried cargo were proposed. Presence within
confined area was also considered as fuzzy set.
Membership function learning method was also
discussed. Arrival and departure from selected routes
crossing areas are trapezoidal imprecise values.
159
Membership functions are to be learned for
particular region, weather condition and each class
of vessels. The imprecise, approximate data were
used to assess navigational situation within routes
crossing area. Product of fuzzy factor and imprecise
weight creates non linear result. To enable
calculations the α-cuts proved to be helpful.
Having at his disposal reliable and fused large
quantity of data and appropriate software tools VTS
operator seems to be able to forecast traffic
congestion within each confined region. Traffic
encountered inside such area is important and
contributes to overall safety standards. Adequate
methods for building hierarchy among alternatives
with widerange of parameter types have been
implemented and discussed by Szlapczynska (Szlap-
czynska 2005).
Multi criteria problem faced by VTS control
station operator was also considered by the author
(Filipowicz 2006). Practical case of decision making
in vessels traffic engineering was presented.
Example included there dealt with system of routes
with rather heavy traffic with one of the vessels that
sought advice on best passage. Best option was
calculated using extended multi-criteria decision aid
software.
REFERENCES
Filipowicz W., Lacki M., Szlapczynska J., 2005. Multicriteria
Decision Support For Vessels Routing, Proceedings of
ESREL’2005.
Filipowicz W., (2006) "Routes Selection and Fuzziness",
Advances in Transport Systems Telematics ed J. Mikulski,
pp. 337-346.
Kaufmann A., Gupta M., 1991. Introduction to fuzzy
arithmetic: theory and application Van Nostrand Reinhold,
New York.
Linn R.J., Hall D.L., “A Survey of Multi-sensor Data Fusion
Systems.” Proceedings of the SPIE - The International
Society for Optical Engineering. 1-2 April 1991: Orlando,
FL. SPIE, 1991. Vol. 1470: (13-29).
Sarma V.S., Raju S., 1991. Multisensor Data Fusion and
Decision Support for Airborne Target Identification, IEEE
Transactions on Systems, Man and Cybernetics. Sept.-Oct.:
pp. 1224-1230.
Szlapczynska J. (2005), „Extended fuzzy TOPSIS – cases of
usage in navigation”, VI-th International Symposium on
Navigation, Gdynia (polish version available).