751
1 INTRODUCTION
With the development of global economy, maritime
transportsystemhasbeenplayinganimportantrole
in the world trading system. However, accidents,
especiallyshipcollisionandgrounding (EMSA,2017),
havebeenimposingathreattosocietyandindividual
in terms of multiple aspects. It is therefore of great
necessitytoconductresearchoncollisionriskanalysis
tofacilitatemaritimesafetyadministrationtoimprove
thesafetylevelandreducetheoccurrenceofcollision
accident.
Toquantitativelyanalysetheriskofshipcollision
accident,variousmethodshavebeenproposed,seeLi
et al (Li et al., 2012). Among them, the
framework
proposed by Fujii (Fujii and Shiobara, 1971) and
Macduff(Macduff,1974)hasbeenwidelyappliedin
regionalcollisionriskanalysisandmanagement.The
frameworkisshowninEq.(1):
=
collisio n geo m etric cau satio n
PP P
(1)
where
g
eom etric
P
denotes the number of collision
candidate,alsoknownasthegeometricalprobability
of collision, which indicates the frequency of ship
encounters that have the potential of collision.
causation
P
indicatestheprobabilityofcollisioncausedby
accident contributing factors, e.g. human and
organisational factors, extreme weather conditions,
and mechanical failures, etc. Such framework
provides a concise approach to estimate the risk of
collision and both the maritime traffic situation and
accident causations can be considered as one
integrity.
Integration of Elliptical Ship Domains and Velocity
Obstacles for Ship Collision Candidate Detection
P.F.Chen&P.H.A.J.M.vanGelder
DelftUniversityofTechnology,Delft,TheNetherlands
J
.M.Mou
WuhanUniversityofTechnology,Wuhan,China
ABSTRACT:Themaritimeshippingindustryhasbeenmakingsignificantcontributionstothedevelopmentof
the regional and global economy. However, maritime accidents and their severe consequences have been
posinganincrementingrisktotheindividualsandsocieties.Itisthereforeimportanttoconductriskanalysis
onsuchaccidentstosupportmaritimesafetymanagement.Inthispaper,amodifiedshipcollisioncandidate
detection method is proposed as a tool for collision risk analysis in ports and waterways. Time‐Discrete
Velocity Obstacle algorithm (TD‐NLVO) is utilized to detect collision candidates based on the encounter
processextracted
fromAISdata.Shipdomainmodelwasfurtherintegratedintothealgorithmasthecriteriafor
determination.Acasestudyisconductedtoillustratetheefficacyoftheimprovedmodel,andacomparison
betweentheexistingmethodandactualshiptrajectoriesarealsoperformed.Theresultsindicatethatwiththe
integration
of ship domain, the new method can effectively detect the encounters with significant collision
avoidancebehaviours.Thechoiceofcriteriacanhaveasignificantinfluenceontheresultsofcollisioncandidate
detection.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 13
Number 4
December 2019
DOI:10.12716/1001.13.04.07
752
To obtain
g
eom etric
P
, generally, there are two major
categoriesofapproaches:1)indicator‐basedapproach
and2)safetyboundaryapproach(Chenetal.,2018).
The indicator‐based approach determines the
encounter situation of ships based on certain
indicators that can reflect their spatiotemporal
proximity, e.g. DCPA (Distance to Closest Point of
Approach), TCPA
(Time to Closest Point of
Approach), relative position, relative speeds and
bearing,etc.Zhangetal.(Zhangetal.,2017)proposed
Vessel Conflict Risk Operator (VCRO) and its
variations facilitate identification of collision
candidateusingAISdata.Lietal(Lietal.,2015)also
utilized the distance between ships,
relative speeds,
coursedifference, etc. to formulate the mathematical
functiontoevaluatetheemergentlevelofencounters.
The safety boundary approach, on the other hand,
determines the encounter situation based on the
violation of certain safety boundary, e.g. Collision
diameter, ship domain, Minimum Distance to
Collision (MDTC) (Montewka et al., 2010),
etc.
Compared with the indicator‐based approach, this
approach considers spatial proximity using the
conceptoftheboundary.FujiiandShiobara(Fujiiand
Shiobara, 1971) first proposed collision diameter as
the boundary to determine which encounter is
dangerous, and such a concept was mathematically
proposed by Pedersen (Pedersen, 1995). Following
such an approach, many similar models have been
developed, see (Ylitalo, 2010). Christian and Kang
(Christian and Kang, 2017) introduced the COWI
model (COWI, 2008) to estimate the probability of
collision of the ship which transports spent nuclear
fuel, and Cucinotta et al (Cucinotta et al., 2017)
utilizedasimilarapproach
toobtainthefrequencyof
ship collision in Messina Strait. Montewka et al
(Montewka et al., 2012). established a probabilistic
model for the marine accident where MDTC is
utilizedascriteriaofcollisioncandidate.Szlapczynski
et al (Szlapczynski and Szlapczynska, 2016)
introduced ship domain as the criteria of collision
candidate
and proposed the degree of domain
violation(DDV)andtimetodomainviolation(TDV)
as indices to reflect the emergent degree of the
encounter.
Althoughvariousmethodshavebeenproposedto
obtain the number of collision candidate, there is
possibilitywhichcouldcauseover/underestimationof
the results. The reason caused such
issues is that
traditional methods do not consider encounter as a
process, instead of the instant information of
encounter,eitherusingindicatororsafetyboundary,
areintroducedtodeterminethesituation.In(Chenet
al.,2018) the authors have changed this perspective,
toconsidertheencounterasaprocessand
determine
collision candidate using Time Discrete Non‐line
Velocity obstacle algorithm (TD‐NLVO). The results
ofthispaperindicatethatcomparedwithtraditional
methods,thenewresultsofthisnewalgorithmshow
highreliability.However,duetothesimplificationin
thiswork,thesafetyboundarywassettobeacircular
shape,whichcouldleadtooverestimationtoa certain
extent.Therefore,inthispaper,thisissueisimproved
withtheintegrationofshipdomainmodel.
In this paper, the previous Time Discrete Non‐
linear Velocity obstacle algorithm is modified with
the integration of ship domain model, to further
improvethe
accuracyoftheresults.Firstly,theNon‐
linearvelocityobstaclealgorithmisintroducedasthe
basic tool to assess encounter situation from the
perspective of the process; Then, the elliptical ship
domainmodelisintegratedintothealgorithmtoact
as criteria of candidate determination. A case study
using actual AIS
(Automatic Information System)
dataisconducted,togetherwithcompassionbetween
the old and new algorithm. The arrangement of the
article is as follows: Section 2 illustrates the
methodologyofthispaper,followedbythedesignof
thealgorithminSection3.Acasestudyisperformed
insection4to
showtheresultsofthealgorithmand
thecomparison.Section5concludesthepaper.
2 METHODOLOGY
According to the definition in (Chen et al., 2018),
collisioncandidateisthepair ofshipsinanencounter
process where their spatiotemporal relationships
satisfy certain criteria that has the potential for
collision.Thisdefinition
providesanopenframework
that can integrate the selected criteria of geometric
collision probability into account. Therefore, in this
paper,theobjectiveistodesignacollisioncandidate
detectionalgorithmthatcandeterminetheencounter
to be dangerous according to the violation of ship
domain of own ship through the
process of the
encounter using historical AIS data in the certain
region.Todoso,TD‐NLVOalgorithmisadoptedas
the basic framework for collision detection, and
elliptical ship domain model is integrated as the
criteria.
3 COLLISIONCANDIDATEDETECTIONMODEL
3.1 TD‐NLVOalgorithm
Velocityobstaclealgorithmisa
typeofalgorithmthat
determinesthepotentialofcollisionbyprojectingthe
spatiotemporalrelationship between own object and
target, e.g. relative position, velocity, etc. into the
velocity space of own object and then checking
whetherownvelocityfallsintothevelocityobstacles
induced by the target. Such methods have been
widely applied in collision detection in robotics
(Fiorini and Shiller, 1998), meanwhile, it is still a
relatively new angle to assess ship collision risk. In
maritime transport field, Degre and Lefevre (Degré
and Lefèvre, 1981) first proposed the idea that
checking the danger of collision using the velocities
between own ship
and target. Such method was
furtherdevelopedandmathematicallyformulatedby
Lenart (Lenart, 1983), which is defined as Collision
ThreatParameterArea (CTPA).Since these methods
assume that the kinematic status of both own ship
andtargetshipremainconstantduringtheencounter,
they are also defined as Linear Velocity
Obstacle
(LVO), which is proved to be identical to CPA
analysisbyHuang,etal(Huangetal.,2017).Dueto
this assumption, the result based on LVO could be
over/estimatedsinceitcannotconsiderthechangesof
both ships’ kinematics during the encounter. To
improvethedeficiencyofLVO,the
constraintofLVO
thatthevelocitiesofshipsremainconstantduringthe
753
encounter is loosen to that velocity of target ship is
flexibleyetknowntoownshipinLargeetal(Largeet
al.,2002).Duringtheencounterprocess,thekinematic
informationofbothshipscanbeupdated,hencetheir
influence on the velocity obstacle induced by the
target.Therefore,in
(Chenetal.,2018)andthiswork
the non‐linear velocity obstacle algorithm was
applied as the fundamental tool for collision
candidate detection. The basic theory of Non‐linear
velocityobstaclealgorithmisshowninFig.1:
Figure1 Basic illustration of Non‐linear Velocity Obstacle
algorithm(Chenetal.,2018)
Suppose that ship A and B in Fig. 1 are in an
encounter situation. The kinematic information of
both ships can be expressed as
AL P t V t
AA A
，，
and
B
LPtVt
BB B
，，
,
respectively.
LPt Vt，，
are their length, position
and velocities at time instance t. Through certain
transformation, suchspatiotemporalrelationshipcan
betransformedintothevelocityspaceofAwiththe
size of A shrinking into a point and B expands to a
largerareaindicatedbyFig.1(b)withradiusRofthe
area.Thisareaisthecollectionofallpotentialposition
of ship A when the collision happens and is also
definedas“conflictposition
ConfP
”(Huangetal.,
2018).The
ConfP
isobtainedaccordingtoEq.(2):
AB
ConfP P t P t R
(2)
Eq. (2) is considered as the criteria of collision
candidate,i.e. ifthedistancebetweentwoshipsfalls
into
ConfP
,collisionisthenlikelytohappenintime
t.consideringthekinematicinformationofownship,
Eq.(2)canberewrittenasEq.(3) ：
AC BC
P
tPt ConfP
(3)
Eq. (3) is an equivalent form of the criteria for
collision candidate. Consider the kinematics of both
ships are known and deterministic and set
AB
VO
as
thevariableofvelocityobstacleofshipAinducedby
targetshipB,Eq.(3)canberewritteninanotherform
asEq.(4)illustrates:
BA0
AB
t
00
() ( )
VO
Pt Pt
ConfP
tt tt
(4)
where
AB
VO
isthesetofvelocitiesofownshipthat
couldleadtocollision.
3.2 Ellipticalshipdomain
In the previous section, the
ConfP
is defined as a
circularareawithradiusR.Suchdefinitionissimilar
to the definition of Collision Diameter proposed by
Fujii(FujiiandTanaka,1971)andPedersen(Pedersen,
1995),henceitalsoinheritsthesimilarissueswhenin
practices: the area is too small that any violation of
suchan
areawouldbephysicalcontact(Montewkaet
al.,2010).Sincecollisioncandidatedenotesthepairof
shipsinencountersituationthat hasthepotential of
the collision, it is reasonable to expand such area to
some extent. In our previous work, the radius was
arbitrarily set to simplify the modelling
complexity,
however, it also brings the issue of potential
overestimation.Inthiswork,weintroducedthestatic
ellipticalshipdomainmodelasthenewcriteria.
Ship domain is firstly introduced by Fujii and
Tanaka(FujiiandTanaka,1971)torepresentanarea
around the ship that would like to keep clear
of
violationofothershipsinthevicinity.Iftheviolation
occurs, it denotes that collision is likely to happen.
Based on this fundamental concept various models
and application have been proposed, e.g.
(Szlapczynski et al., 2018; Wang, 2010), etc. In this
paper, we replace the circular
ConfP
with a static
ellipticalshipdomainasthenewcriteriaforcollision
candidate. The parameters of such area are semi‐
major and semi‐minor axis, respectively. In this
paper, they are set as 8 times and 4 times of own
ship’slengthbasedontheresearchby(Szlapczynski
et al., 2018).
To integrate such domain into the TD‐
NLVO, a mathematical function needs to be
proposed. To do so, two va riables of own ships
informationneedstobeintegrated,whicharelength
and course over ground, respectively. Such
parameterscanbeobtainedinhistoricalAIS data.The
generalfunctionofan
ellipsecanbewrittenasEq.(6):
22
0Ax Bxy Cy Dx Ey F
(5)
ItsidenticalformisshowninEq.(7):
22
22
1
xy
ab
(6)
Eq.(6)andEq.(7)describes theellipsewhosefoci
areeither on the major and minor axis. In practices,
such domain needs to be described in the local
coordinatesofownshipwhosemajoraxisisalongthe
course, instead of true north. Therefore, the ship
domain needs
to be rotated according to the course
information.TherotationfunctionisshowninEq.(8):
'
'
cos sin
sin cos
xx y
yx y
(7)
Therefore,therotatedshipdomaincanbewritten
asEq.(9):
22
22
cos sin sin cos
1
xy xy
ab
(8)
754
Rewriting Eq. (9) according to Eq. (1), the
parameterfunctionisshowninEq.(10):
22 2 2 2 2 2 22 2
22 22
sin cos cos sin
2sincos 0
ab xab y
ab xyab
(9)
Thenthecorrespondingparametersoftherotated
shipdomaincanbeobtainedaccordingtoEq.(11):
22 2 2
22
22 22
22
sin cos
2sincos
cos sin
Aa b
Bab
Ca b
F
ab
(10)
3.3 Designofcollisioncandidatedetectionmodel
Withtheintegrationof TD‐NLVOandellipticalship
domainmodel,thenewversionofshipdomain‐based
collision candidate detection model can be
established.Asaforementioned,collisioncandidateis
detected according to the total process of encounter,
insteadofinstanceencounter
informationatacertain
time interval. To do this, how to construct the
trajectory data and process them are one of the
importanttechnicalproblemshere.Toimplementthe
process perspective, the historical AIS data of ships
navigating in the area are first reconstructed as
chronologicaltrajectorydataaccordingtotheir
MMSI
(MaritimeMobileServiceIdentifier).Toacceleratethe
computing speed, such a trajectory is also divided
into subsets using the same parameter settings in
(Chen et al., 2018). The design of the new model is
showninFig.(2):
Figure2. Flow chart of the ship domain‐based collision
candidatedetectionmodel
4 CASESTUDY
In this section, a case study on implementing the
domain‐basedTD‐NLVOisillustrated.TheAISdata
are obtained from the open access provided by the
Danish Maritime Authority. Since the goal of this
paper is to demonstrate the efficacy of the modified
TD‐NLVO and the
comparison between the original
andnewmethod.Thetimespanofthedataissetto
be1day.HereweintroducedtheAISdataon1
st
Oct.
2018 in port Aarhus as the test datasets. The
parameter settings are as follows:
Scan
T
is set to be
60minsand
theshold
T
:30s;
The encounter between tanker “219XXX000” and
cargo ship “257XXX000” are shown with their
trajectoryandencountersituationinvelocityspaceat
a certain time step. Based on the AIS data and TD‐
NLVO, these two ships have an encounter that
violatesthedomainofownship.Thetrajectoriesare
showninFig.3:
Figure3.Trajectorybetweentwoships
Fromthereconstructofshiptrajectorieswecansee
thatatthebeginningoftheencounterprocess,tanker
“219XXX000” and cargo ship “257XXX000” were in
“head on” situation. With both ships approaching
eachother,ownship(blue)detectedthattheremight
be danger of collision, therefore, she altered her
course to
her starboard to enlarge the distance
between both ships, meanwhile the cargo ship also
alteredhercoursetoherstarboardsideabittomake
surebothcanpasseachother on herstarboardside,
which is required by the COLREGs. Taking the
encounter situation at 11:41:35 AM as the
example,
the spatiotemporal relationships between both ships
invelocityspaceofownshipisshowninFig.(4):
Figure4.Encountersituationinvelocityspaceofownship
755
Figure5 Crossing encounter situation between
“219XXX172”and“219XXX903”
AsFig.4indicates,theshipdomain‐basedvelocity
obstacle algorithm successfully detect a violation of
shipdomaininthefutureofdetectiontime11:41:35,
i.e. at that time, TD‐NLVO detected that the ship
domainofownshipwouldbeviolatedinthefuture.
Compared with trajectory information in Fig.
4, the
alertfromVOisearlierthantheactualmovement.
Fig. 5 illustrates another encounter situation
between “219XXX172” and “219XXX903” at the
entrance of port Aarhus. With the use of domain‐
basedTD‐NLVO,theencounterprocesscanbeeasily
demonstrated in velocity space of own ship. As we
can see, at 8:43:33 AM when own ship went
outbound, with trajectory information of target ship
she can detect the violation of velocity obstacle at a
certain point in the future at that time. With the
developmentoftime,ownshipchosetomakeaturn
to her starboard to follow
the channel and avoid
possiblecollision with target ship. Besides, since the
modifiedTD‐NLVOconsideredcourseinformationin
domainmodelling,thecoverageofvelocityobstacles
are different during the encounter process as the
courseofownshipchangesconstantly.
5 DISCUSSIONS
Inthissection,acomparisonbetweentheoriginal
TD‐
NLVO(M1)andshipdomain‐basedTD‐NLVO(M2)
isconducted.Thecomparisonhastwocomponents:1)
comparisonbetweenresultsfromM1andM2and2)
detail analysis of the common results from two
differentmethods.TheAISdatautilizedishistorical
AISdataofportArhuson1
st
Oct2018fromtheopen
access of the Danish Maritime Authority. The
parameter setting between the two methods are as
follows:
Table1.ParametersettingsforM1andM2
_______________________________________________
Parameter M1M2
_______________________________________________
Ttheshold30s30s
T
scan60min60min
criteriaCircularsafety Ellipticalshipdomain
region
radius1000msemi‐major:8*length(m)
semi‐minor:4*length(m)
_______________________________________________
Withtheapplicationoftwomethodsonthesame
AIS data, the results are significantly different: for
M1, the number of collision candidate is 19
meanwhile the number of collision candidate
obtainedwithM2is7.Thedetailoftheresultscanbe
foundin Table 1, Appendix I. The difference
can be
explained by the difference in criteria choices. As
proved in our previous work (Chen et al., 2018),
differencechoicesonthecriteriaofcollisioncandidate
may reveal significant results. However, compared
with circular shape region, ship domain, due to its
capabilityonexpressingthepreferenceofcoverageon
different azimuth around ship according to various
aspects, e.g. experience of the officer on watch, ship
manoeuvrability, etc. is reasonable to be integrated
into TD‐NLVO. From the tra jectory data of the
collisioncandidate,wehavefoundoutthatM1have
detected some encounters that do not have obvious
collisionavoidance
behaviour, e.g.encounterbetween
756
ship“219XXX172”and“219XXX000”whichisshown
inFig.8(a).Forthe12encountersthatwereidentified
byM1butignoredbyM2,8outofthemfallsintothis
type while rest of them shown certain avoidance
behaviour which was ignored by M2, e.g. encounter
between“219XXX000”and
“248XXX000”(Fig.6(b)).
a)
b)
Figure6.IllustrationofencounterobtainedwithM1
From the data analysis, we can see that the
determinationofcriteriaforcollisioncandidatehavea
stronginfluenceontheabsolutenumber.Inpractices,
suchcriteriashouldbedeterminedwithcautionand
taking region traffic characteristics, ship
characteristics,etc.intoconsideration.
As for the 7 common collision candidates from
both
methods,onecanfindthatthestartofdetection
anddurationofviolationforresultsobtainedwithM1
isinadvancetotheresultsobtainedwithM2tosome
extent. The detailed information can be found in
Table 2, Appendix I. Taking the encounter between
“219XXX000” and “257XXX000” as an
example. The
trajectoriesofthetwoshipsareshowninFig.3.The
velocity obstacle utilized in M1 and M2 at time
instance “11:41:35 AM” are shown in the figure
below:
a)
b)
Figure7.IllustrationofVelocityobstacleobtainedwithM1
andM2at11:41:35AM
From Fig. 7 we can see that at the time step the
velocityobstacle obtained withthe circular region is
obviously larger than ship domain‐based velocity
obstacle.Twoaspectscanexplainthis:1)theradiusof
thecircularregionissettobe1,000mwhiletheship
domain of
own ship “219XXX000”(85 meters in
length) is smaller than such value; 2) due to the
introduction of ship domain, the course information
ofownshipisalsoconsideredinthemodelasshown
in Fig. 5. With the different course of own ship, the
coverageof velocityobstacle is also
different.Under
theinfluenceofthesetwofactors,wecanseethatfor
Fig.9(b)thevelocityofownshipalreadyfallsintothe
VOwhileinFig.7(a)itonlyfallsattheboundaryof
the VO, which can be utilized to explain the time
advanceintheresults
fromM1.
6 CONCLUSIONS
The maritime transport system is an important
componentoftheglobaltransportationsystem.Inthis
paper, a modified non‐linear velocity obstacle
algorithm integrating elliptical ship domain is
proposed. Elliptical ship domain was integrated as
the new criteria of collision candidate. Based on the
historicalAIS
data,acasestudywasimplementedto
demonstrate the efficacy and comparison between
existing VO algorithm. The results indicate that: 1)
with the integration of elliptical ship domain, the
modifiedTD‐NLVO algorithm can reflect the course
changes in velocity space of own ship during the
encounter;2)comparedtothe
originalversionofthe
algorithm,themodifiedversionreducedthechances
of false positive detection of the encounters that do
not have obvious collision avoidance behaviours,
757
however, it also leads to a certain extent of false
negativeresults;3)comparedtocriteriaofthesafety
region, the detection time and start of violation of
domain‐basedmodelarelatertosomeextent,which
is caused by the reduced coverage of velocity
obstacle.
Based on the results,
one can see that with the
integrationofshipdomainmoreinformation(course
and length of ships) can be incorporated into the
processofcollisioncandidatedetection.However,the
choice of parameters, such as the parameters of the
shipdomain,haveastronginfluenceontheabsolute
number of collision
candidates detected by the
methods.Therefore, to furtherimprove the accuracy
of the methods, further efforts can be devoted to
determiningthecriteriaconsideringcharacteristicsof
regional traffic, e.g. distribution of ship length, etc.
Another aspect that needs further work is how to
improve the quality of data to avoid potential
underestimation of the results since in the data we
introduced we have identified multiple cases where
shiplengthinformationismissing.Infutureresearch,
thedomainofbothshipswillbetakenintoaccountto
avoid the situation where two ships in encounter
situationdetectsviolationatdifferenttime.
ACKNOWLEDGEMENT
This work is supported by the China Scholarship Council
under Grant: 201606950005 and the National Science
Foundation of China (NSFC) through Grant No.51579201.
ThehistoricalAISdataisprovidedbytheopenaccessofthe
Danish Maritime Authority. The corresponding author
would also like to thank Sihui Hu for her encouragement
alongourjourney.
APPENDIXICOLLISIONCANDIDATEINFORMATIONOBTAINEDFROMBOTHMODELS
Table1.Resultsofcollisioncandidateobtainedfrombothmodels
__________________________________________________________________________________________________
Model Ownship DetectionperiodTargetship Duration
__________________________________________________________________________________________________
M1 219XXX000 06:40:5306:46:13220XXX000 06:42:5106:48:22
219XXX313 07:31:2707:34:26219XXX172 07:31:3007:34:48
219XXX72 07:16:5707:17:45219XXX000 07:17:3707:17:49
219XXX172 08:43:2308:47:39219XXX903 08:46:1908:47:42
219XXX172 11:30:5511:39:30219XXX000
11:34:4211:39:40
219XXX172 11:26:5711:27:43257XXX000 11:27:0311:28:00
219XXX000 11:42:2511:59:34248XXX000 11:59:2312:00:57
219XXX000 11:39:5511:47:15257XXX000 11:45:5411:47:39
219XXX000 12:07:3312:11:31219XXX000 12:11:1912:11:44
219XXX000 12:04:3712:06:49248
XXX000 12:06:2812:06:58
219XXX000 12:00:1912:01:09257XXX000 12:00:2512:01:50
219XXX000 12:00:0412:00:38248XXX000 12:00:1212:00:57
219XXX172 13:06:2113:11:13256XXX000 13:10:4213:11:27
219XXX903 14:00:0214:01:52246XXX000 14:00:4214:02:08
211XXX340 15:18:06
15:47:36219XXX307 15:44:2016:03:56
219XXX172 15:42:3915:43:01219XXX000 15:42:4515:43:05
212XXX000 16:45:5116:56:09219XXX903 16:55:3116:56:20
219XXX172 17:00:1817:03:45219XXX903 17:02:4417:03:48
209XXX000 18:00:0918:20:30219XXX000 18:19:1518:31:36
M2 219XXX000 06:42:5506:46:13220XXX000 06:44:5606:48:22
219XXX172 08:43:2308:47:39219XXX903 08:46:4508:47:42
219XXX172 11:33:0711:39:30219XXX000 11:38:2011:39:40
219XXX000 11:41:3511:47:15257XXX000 11:46:3411:47:39
219XXX903 14:01:1814:01:52246XXX000
14:01:2614:02:08
219XXX172 17:00:1817:03:45219XXX903 17:02:5517:03:48
209XXX000 18:00:0918:20:30219XXX000 18:19:2118:31:36
__________________________________________________________________________________________________
Table2.Detailsofcommoncollisioncandidatefrombothmodels
__________________________________________________________________________________________________
Model Ownship DetectionperiodTargetship Duration
__________________________________________________________________________________________________
M1 219XXX000 06:40:5306:46:13220XXX000 06:42:5106:48:22
219XXX172 08:43:2308:47:39219XXX903 08:46:1908:47:42
219XXX172 11:30:5511:39:30219XXX000 11:34:4211:39:40
219XXX000 11:39:5511:47:15257XXX000 11:45:5411:47:39
219XXX903 14:00:0214:01:52246XXX000
14:00:4214:02:08
219XXX172 17:00:1817:03:45219XXX903 17:02:4417:03:48
209XXX000 18:00:0918:20:30219XXX000 18:19:1518:31:36
M2 219XXX000 06:42:5506:46:13220XXX000 06:44:5606:48:22
219XXX172 08:43:2308:47:39219XXX903 08:46:4508:47:42
219XXX172 11:33:0711:39:30219XXX000 11:38:2011:39:40
219XXX000 11:41:3511:47:15257XXX000 11:46:3411:47:39
219XXX903 14:01:1814:01:52246XXX000
14:01:2614:02:08
219XXX172 17:00:1817:03:45219XXX903 17:02:5517:03:48
209XXX000 18:00:0918:20:30219XXX000 18:19:2118:31:36
__________________________________________________________________________________________________
758
REFERENCE
Chen,P.,Huang,Y.,Mou,J.,vanGelder,P.H.A.J.M.,2018.
Ship collision candidate detection method: A velocity
obstacleapproach.OceanEngineering170,186‐198.
Christian,R., Kang,H.G.,2017.Probabilisticriskassessment
on maritime spent nuclear fuel transportation (Part II:
Ship collision probability). Reliability Engineering &
SystemSafety164,136‐149.
COWI, 2008. Risk Analysis Sea trafic Area around
Bornholm.
Cucinotta,F.,Guglielmino,E.,Sfravara,F.,2017.Frequency
of Ship Collisions in the Strait of Messina through
RegulatoryandEnvironmental ConstraintsAssessment.
JournalofNavigation70(5),1002‐1022.
Degré,T.,Lefèvre,X.,1981.A CollisionAvoidanceSystem.
JournalofNavigation
34(02),294‐302.
Fiorini, P., Shiller, Z., 1998. Motion planning in dynamic
environmentsusingvelocityobstacles.TheInternational
JournalofRoboticsResearch17(7),760‐772.
Fujii, Y., Shiobara, R., 1971. The Analysis of Traffic
Accidents.JournalofNavigation24(04),534‐543.
Fujii, Y., Tanaka, K., 1971. Traffic Capacity.
Journal of
Navigation24(04),543‐552.
Huang, Y., van Gelder, P., Mendel, M.B., 2017. Imminent
ships collision risk assessment based on velocity
obstacle.
Huang,Y., vanGelder,P.H.A.J.M.,Wen,Y.,2018.Velocity
obstacle algorithms for collision prevention at sea.
OceanEngineering151,308‐321.
Large, F., Sekhavat, S., Shiller,
Z., Laugier, C., 2002.
Towardsreal‐timeglobalmotionplanninginadynamic
environment using the NLVO concept, IEEE/RSJ
International Conference on Intelligent Robots and
Systems,pp.607‐612vol.601.
Lenart, A.S., 1983. Collision Threat Parameters for a New
Radar Display and Plot Technique. Journal of
Navigation36(3),404‐410.
Li, S., Meng, Q., Qu, X., 2012. An overview of maritime
waterway quantitative risk assessment models. Risk
Anal32(3),496‐512.
Li, S., Zhou, J.H., Zhang, Y.Q., 2015. Research of Vessel
Traffic Safety in Ship Routeing Precautionary Areas
Based on Navigational Traffic Conflict Technique.
JournalofNavigation68(3),
589‐601.
Macduff,T.,1974.Theprobabilityofvesselcollisions.Ocean
Industry9(9).
Montewka,J.,Goerlandt,F.,Kujala,P.,2012.Determination
ofcollisioncriteriaandcausationfactorsappropriateto
a model for estimating the probability of maritime
accidents.OceanEngineering40,50‐61.
Montewka, J., Hinz, T., Kujala, P., Matusiak,
J., 2010.
Probability modelling of vessel collisions. Reliability
Engineering&SystemSafety95(5),573‐589.
Pedersen, P.T., 1995. Collision and grounding mechanics.
ProceedingsofWEMT95(1995),125‐157.
Szlapczynski, R., Krata, P., Szlapczynska, J., 2018. Ship
domain applied to determining distances for collision
avoidance manoeuvres in give‐way
situations. Ocean
Engineering165,43‐54.
Szlapczynski, R., Szlapczynska, J., 2016. An analysis of
domain‐based ship collision risk parameters. Ocean
Engineering126,47‐56.
Wang,N.,2010.AnIntelligentSpatialCollisionRiskBased
ontheQuaternion ShipDomain.JournalofNavigation
63(4),733‐749.
Ylitalo,J.,2010.Modellingmarine
accidentfrequency,Alto
UniversitySchoolof ScenceandTechnologyFacultyof
InformationandNaturalScience.
Zhang, W.B., Kopca, C., Tang, J.J., Ma, D.F., Wang, Y.H.,
2017.A SystematicApproachforCollisionRiskAnalysis
based on AIS Data. Journal of Navigation 70 (5), 1117‐
1132.
