Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Collaborated and Constrained Neural-EKF Algorithm for the Vessel Traffic Monitoring and Information System

Collaborated and Constrained Neural-EKF Algorithm for the Vessel Traffic Monitoring and Information System

Ratings: (0)|Views: 15 |Likes:
Published by LP Perera
Maneuvering vessel detection and tracking in cooperation
with vessel state estimation and navigational trajectory
prediction are important tasks for the Vessel Traffic Monitoring
and Information Systems (VTMIS) to improve maritime safety
and security in ocean navigation. In this study, collaborated
and constrained Neural-EKF algorithm is proposed for the
above purpose. The proposed methodology consists of two main
units: an Artificial Neural Network based Vessel Detection and
Tracking Unit and an Extended Kalman Filter based State
Estimation and Trajectory Prediction Unit. Finally, the
proposed algorithm, is implemented on the MATLAB software
platform, and successfully illustrate the results attainable in
respect to vessel detection and tracking, vessel state estimation
and navigational trajectory prediction in ocean navigation is
also presented in this study.
Maneuvering vessel detection and tracking in cooperation
with vessel state estimation and navigational trajectory
prediction are important tasks for the Vessel Traffic Monitoring
and Information Systems (VTMIS) to improve maritime safety
and security in ocean navigation. In this study, collaborated
and constrained Neural-EKF algorithm is proposed for the
above purpose. The proposed methodology consists of two main
units: an Artificial Neural Network based Vessel Detection and
Tracking Unit and an Extended Kalman Filter based State
Estimation and Trajectory Prediction Unit. Finally, the
proposed algorithm, is implemented on the MATLAB software
platform, and successfully illustrate the results attainable in
respect to vessel detection and tracking, vessel state estimation
and navigational trajectory prediction in ocean navigation is
also presented in this study.

More info:

Categories:Types, Research
Published by: LP Perera on Jun 22, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

10/25/2012

pdf

text

original

 
1 Copyright © 2011 by ASME
Proceedings of the 30
th
International Conference on Ocean, Offshore and Arctic EngineeringOMAE2011June 19-24, 2011, Rotterdam, The Netherlands
OMAE2011-50248
COLLABORATED AND CONSTRAINED NEURAL-EKF ALGORITHM FOR THEVESSEL TRAFFIC MONITORING AND INFORMATION SYSTEM
Lokukaluge P. Perera
Centre for Marine Technology and Engineering (CENTEC),Technical University of Lisbon, Instituto Superior Técnico,Av. Rovisco Pais,1049-001 Lisboa, PortugalEmail : prasad.perera@mar.ist.utl.pt
Paulo Oliveira
Institute for Systems and Robotics (ISR),Technical University of Lisbon, Instituto Superior Técnico,Av. Rovisco Pais,1049-001 Lisboa, PortugalEmail : pjcro@isr.ist.utl.pt
C. Guedes Soares
Centre for Marine Technology and Engineering (CENTEC),Technical University of Lisbon, Instituto Superior Técnico,Av. Rovisco Pais,1049-001 Lisboa, PortugalEmail : guedess@mar.ist.utl.pt
ABSTRACT
 Maneuvering vessel detection and tracking in cooperationwith vessel state estimation and navigational trajectory prediction are important tasks for the Vessel Traffic Monitoringand Information Systems (VTMIS) to improve maritime safetyand security in ocean navigation. In this study, collaborated and constrained Neural-EKF algorithm is proposed for theabove purpose. The proposed methodology consists of two mainunits: an Artificial Neural Network based Vessel Detection and Tracking Unit and an Extended Kalman Filter based State Estimation and Trajectory Prediction Unit. Finally, the proposed algorithm, is implemented on the MATLAB software platform, and successfully illustrate the results attainable inrespect to vessel detection and tracking, vessel state estimationand navigational trajectory prediction in ocean navigation isalso presented in this study.
NOMENCLATURE
R
L1
Neural network constrained learning rangeR
L2
 
The sensor maximum measurement ranger
 j
(k) j-th position range vectorr(k)
 
Accumulated range vector
(k)
 j
ϑ 
  j-th position bearing vector
(k)
ϑ 
 Accumulated bearing vector
c
x
 j
(k),
c
y
 j
(k) j-th position x, y coordinatesx
 j
(k),y
 j
(k),z
 j
(k) Normalized j-th position x, y, z coordinatesp(k) Accumulated normalized position vectorp
 j
(k) j-th position vectorW(k) Accumulated prototypes vectorw
i
(k)
 
i-th prototype vectorsn(k) Neural network inputa(k) Neural network output vector
θ
i
(k)Angle between
 
p
 j
(k) and w
i
(k)
α
 
Learning rate of the Instar rulex(t), y(t)
 
Vessel position x, y coordinatesv
x
(t), v
y
(t)
 
Vessel x, y directional velocity componentsa
n
(t), a
t
(t) Vessel normal, tangential accelerationV
a
(t) Vessel speed
χ 
a
(t)
 
Vessel coursex(t) Nonlinear vessel state vectorf (x(t)) Nonlinear vessel state functionw(t)~N(0,Q(t)) Vessel state noise vectorz(t) Measurement vectorz
x
(k), z
y
(k) Measurement x, y vessel coordinatesh(x(t)) Measurement functionv(t)~N(0,R(t)) Measurement noise vectorQ(t) Vessel state noise covarianceR(k) Measurement noise covariancex(t) Nonlinear vessel state vector
)t(x ~
 Vessel state error vector)t(x ˆ Estimated vessel state vectorx(k 
-
), x(k 
+
) Estimated prior and posterior state vectorsf (x(t)) Nonlinear vessel state function
 
2 Copyright © 2011 by ASMEw(t) Vessel state noise vectorz(t) Measurement vectorh(x(k)) Measurement functionv(k) Measurement noise vectorQ(t) Vessel state noise covarianceR(k) Measurement noise covarianceP(t) Estimated error covarianceP(k 
-
), P(k 
+
) Estimated prior and posterior errorcovariance of state vectorsK(k) Kalman filter gain
INTRODUCTION
The European Union (EU) is surrounded by one of thebusiest and most complex sea route systems in the world. Over90% of the EU external trade goes by sea and over 3.7 billiontones of the freight per year are transferred through the EUports. In addition, passenger traffic in the seas around the EUregions is presently approximated to 350 million passenger journeys per year [1].With the increase demand for maritime transportation of passengers and freights in the EU maritime regions, the safetyand security issues in ocean navigation are highlighted.Furthermore, a maritime monitoring mechanism for thatpurpose is proposed by the EU Directive 2002/59 [2], where thehighly dense maritime traffic regions are to be equipped withthe regional Vessel Traffic Monitoring and Information Systems(VTMISs). The conventional ocean navigational systems andVTMISs are equipped with several maritime surveillancesystems (i.e. Radar, Laser, Ladar, Automatic Radar Plotting Aid(ARPA), Automatic Identification System (AIS), and Long-Range Identification and Tracking System (LRIT)) for the samepurpose. However, there are many challenges faced by thosemaritime surveillance systems as presented in [3]: The largersurveillance volume, synchronization among targets andsensors, noisy signal propagation environment and multi-targetobservations. Therefore, the integration of advanced featuresinto the maritime surveillance systems as proposed in this studyare required to overcome the challenges faced in oceannavigation.However, there are some advanced features developedunder current maritime surveillance systems of ARPA and AIS.The ARPA system provides accurate information on range andbearing of nearby vessels and the AIS is capable of givinginformation on the vessel structural data, position, course, andspeed, etc. The AIS marine traffic simulator, aiming to performnavigation safety and security studies, is presented in [4]. Eventhough the ARPA and AIS systems were developed to providenavigation aids (ie. detection and tracking facilities) to vessels,the vessel state estimation and its navigational trajectoryprediction tasks are still under developed.The main advantages of vessel state estimation andnavigational trajectory prediction tasks are the intention of thevessel and the collision risk among vessels to be predictedahead of time. Furthermore, those tasks are also important toolsto Long-Range Identification and Tracking (LRIT) system, thatwill be implemented as an international maritime securitynetwork, where each respective country has sufficient time toevaluate and to response to security risk that is posted by eachvessel to its coastline [2]. Furthermore, by obeying to theconditions of national and international maritime laws, by eachvessel, can also be monitored by these systems where thecoastal security and safety issues should not be compromised.The VTMIS proposed in this work is presented in Figure 1.As depicted in the figure, the system consists of two main units:ANN (Artificial Neural Network) based Vessel Detection unitand Tracking Unit and EKF (Extended Kalman Filter) basedTrajectory Estimation and Prediction Unit. The combination of these units, collaborated and constrained Neural-EKF algorithmthat is the main objective in this paper, is described in thefollowing sections.The work presented in this study is a part of the ongoingeffort to formulate an Intelligent Collision Avoidance System inocean navigation, as further described by [5] and [6]. Theorganization of this paper is as follows. The second sectioncontains an overview of proposed Vessel Traffic Monitoringand Information System (VTMIS). An ANN based vesseldetection and tracking process and an EKF based stateestimation and trajectory prediction process are presented in thethird and forth sections respectively. The fifth section contains adetailed description of computational simulations anddiscussions. Finally, the conclusion and future work arepresented in the sixth section.
Figure 1. The Proposed VTMIS
 
3 Copyright © 2011 by ASME
VESSEL TRAFFIC MONITORING AND INFORMATIONSYSTEM
The mathematical formulation of the VTMIS that ispresented in Figure 1 is further discussed in this section. Amulti-vessel situation under the radar/laser sensor measurementsis presented in Figure 2. The radar/laser sensor is located inposition O (0, 0) and the i-th vessel at the k-th time instant islocated at the position A
i
(k). As presented in the figure, eachvessel is identified as a cluster of data points that is observed bythe radar/laser sensor. The sensor range is divided into tworegion with respect to radius of R
L2
and R
L1
as presented in thefigure.In the region between regions R
L2
and R
L1
the vesseldetection and tracking processes are executed. In the regionR
L1
the vessel states (i.e. position, velocity and acceleration)estimation and navigational trajectory prediction processes areexecuted. Furthermore, the initial neurons that are used for thevessel detection and tracking process are located in theboundary layer region of R
L2
as further described in the thirdsection.
Sensor Measurement and Coordinate Systems
After radar/laser scan, the sensor generates thecorresponding range and bearing values of the vessels andobstacles in the environment, at k-th time instant can be writtenas:
[ ][ ]
)(...)()()(  )(r...)(r)(r)(r
R21R21
ϑϑϑ=ϑ =
(1)The range and bearing values are transformed into theCartesian coordinated in the j-th position at the k-th time instantcan be written as:
( )( )
)(sin)(r)(y  )(cos)(r)(x
 j j jc j j jc
ϑ=ϑ=
(2)However, these position data points should be normalizedwith respect to the maximum range of the radar/laser sensor.The normalized j-th position coordinates of the vessel at the k-th time instant can be written as:
2L jc j2L jc j
R)(y )(y  R)(x )(x
==
(3)However, for the fair neural competition these positionvectors should be formulated for unit magnitude conditions asfurther described in the following section, where the thirddimension, z
i
(k), is introduced, and that can be written as:
1)(z)(y)(x
2 j2 j2 j
=++
(4)The coordinate of the z
i
(k) can be calculated consideringthe above unit magnitude conditions for the j-th data point andcan be written as:
)(y)(x1)(z
22
 j j j
=
(5)One should note that the introduction of a third dimension(z
i
(k)) can be evaluated as a transformation of 2D positioncoordinates into 3D position coordinates. The complete 3D j-thposition data point at the k-th time instant can be written as:
)(z)(y)(x)(p
 j j j j
(6)Therefore, the complete data cluster (see Figure 3) that isgenerated by the sensor measurements can formulate as:
[ ]
)(p)(p)(p)(p
R21
K
(7)
ANN BASED VESSEL DETECTION AND TRACKING
The theoretical foundation of artificial neurons isformulated from observations of biological concepts as well asinspirations of the behavior of human brain and nervous system.An artificial neuron consists of several inputs that correspond tothe synapses of a biological neuron. Furthermore, it consists of only one output that corresponds to the axon of a biologicalneuron. However, each input corresponds to a certain weightvalue that influences the corresponding signal over the neuronoutput.
Figure 2. Multi-vessel Radar/Laser Measurement

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->