301
1 INTRODUCTION
Maritime traffic plays a substantial role in global trade.
More than 80% of world trade volume is transported
by sea [1]. In coastal waters the maritime traffic density
is particularly high. To ensure safety and efficiency of
the maritime traffic, so-called Vessel Traffic Service
(VTS) has been established [2]. VTS centers world-wide
monitor and coordinate maritime traffic. To do so they
rely on various communication and sensor
technologies. The International Association of Marine
Aids to Navigation and Lighthouse Authorities (IALA)
provides guideline concerning technical requirements
for VTS systems [3]. In Tab. 1 the most common
communication and sensor technologies are
summarized. Based on the given technologies a Vessel
Traffic Service Operator (VTSO) can obtain situational
awareness of his designated VTS territory.
To enhance the situational awareness of the
maritime traffic the IALA motivates VTSOs to use
LEAS: An AI-based Demonstrator as Decision Support
Tool for Traffic Monitoring at VTS Centers
T. Stach & P. Koch
Fraunhofer Centre for Maritime Logistics and Services CML, Hamburg, Germany
ABSTRACT: Maritime traffic is prevalent worldwide, with particularly high density in coastal waters. To ensure
safety and efficiency, Vessel Traffic Service (VTS) centers monitor and coordinate maritime traffic. For this
purpose, VTS centers utilize various sensor and communication technologies such as radar, Automatic
Identification System (AIS), electro-optical systems or radio communication. Additionally, any Vessel Traffic
Service Operator (VTSO) is motivated to utilize a Decision Support Tool (DST). The LEAS project addresses
emerging challenges at VTS centers. One key challenge results from the continuous evolution of maritime traffic,
in particular, its ever increasing automation and autonomization. Another key challenge is the growing shortage
of skilled workers. Consequently, it is crucial to process increasing volume of maritime traffic data while
maintaining or improving safety and efficiency. DSTs at VTS centers must be adapted to these emerging
challenges, accordingly. In the LEAS project, we develop and evaluate a demonstrator which represents a DST.
This demonstrator is being developed in close collaboration with VTSOs to address these challenges. Most
notably, it has a situation detection which makes use of Artificial Intelligence (AI) methods and displays relevant
information in an intuitive Human-Machine Interface (HMI). The demonstrator is evaluated using simulated
traffic scenarios in the German Bight and Baltic Sea, with VTSOs as test subjects. This paper provides an overview
of the project and demonstrator. First, we introduce the key requirements for the demonstrator and discuss their
impact on the system architecture. Next, we present its AI-based situation detection. We explain the underlying
formalism of the situation detection and resolution as well as its implementation in the demonstrator. Finally, we
evaluate the capabilities and limitations. The paper concludes with an outlook to future work with focus on
potential deployment at DST at VTS centers.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 19
Number 1
March 2025
DOI: 10.12716/1001.19.01.35
302
Decision Support Tools (DSTs) [4]. Accordingly, a DST
should support the VTSO by various means:
− “provide alerts and indicators”
− “reduce the workload”
− “enhance efficiency”
− “be accurate and in real time”
Table 1. Communication and sensor technologies of VTS
centers as defined by IALA in Guideline G1111 [3].
Technology
Radar
AIS
Environmental
monitoring
Electro-optical
systems
Radio
communications
Radio direction
finders
As pointed out in a previous article, current VTS
systems and DSTs have many technical limitations [5],
[6]. VTS centers collect a high volume of multimodal
data through communication and sensor technologies
(cf. Tab. 1). However, there is no known VTS system or
DST which fuses the multimodal data stream and
creates a holistic situational awareness. For example,
AIS data could be fused with VHF radio
communication to provide more contextual
information on arising situations.
Instead VTSO rely on manual tools such as creation
of threshold-based, spatio-temporal alerts. From
differences in regulatory and physical properties of
VTS territories, differing traffic patterns result. Thus,
substantial experience is mandatory with respect to
each VTS territory. Also, here Machine Learning (ML)-
based techniques could have been by applied. For
example, automatized statistical analyses and
extraction of traffic parameters can represent human
experience to some degree.
The urgency for novel DSTs is exacerbated by
external factors. On the one hand, VTS is facing
constant evolution of maritime traffic, in particular,
increasing automatization and autonomization. On the
other hand, the shortage of skilled workers at VTS
centers is increasing as stated by cooperating VTS
centers in Travemünde and Warnemünde and incident
investigation reports [7]. This creates a growing gap
between the ever-changing monitoring and
coordination tasks and available skilled workers. The
growth of this gap impairs the safety and efficiency of
maritime traffic. Already today, there are avoidable
incidents [7].
The project LEAS aims at filling this gap. The
outcome of the project is a demonstrator which
represents a next- generation DST for VTS centers.
Through the integration of state-of-the-art
technologies, such as ML-based data processing and
fusion, the demonstrator aims at giving a deeper and
more holistic situational awareness. This is primarily
achieved by the Artificial Intelligence (AI)-based
situation detection and resolution module which is
presented in this paper. Furthermore, the utilized
techniques facilitate scalability to multiple VTS
territories and extensibility of integratable data streams
and detectable situations. Due to the implemented
formalism not only situations are adaptable and
intuitively interpretable. Also, situations are
automatically combined with each other. This lowers
the number of false alerts and cognitive workload.
Thus, the focus on true alerts is improved.
The remainder of this article is organized as follows.
In Ch. 2 related works are presented and compared to
our work. Then, the superordinate project of this work
is presented in Ch. 3. Subsequently, in Ch. 4 our
demonstrator is discussed in detail. This chapter
represents the main part of this article and is divided
into several sub-chapters. Finally, this paper ends with
a conclusion of the results to date and gives an outlook
to future work.
2 RELATED WORKS
An extensive survey on (anomalous) situation for
maritime traffic monitoring purposes has been carried
out by us in a preceding work [8]. Thus, here we limit
related works to articles published from year 2020
onwards to ensure novel approaches. Also, the focus
lies on rather holistic approaches and advanced
technological readiness than highly specific solutions.
Xiao et al. provide requirements for a framework
for a next- generation VTS system [6]. According to the
authors, it should be equipped with AI-based
functionalities such as forecast of traffic-dense hot
spots, creation of knowledge bases (based on pattern
extractions) and the incorporation of Electronic
Nautical Charts (ENC) into automated decision
support. The next-generation VTS system should aim
at further automating maritime traffic management,
inter alia, by reducing the involvement of a human in
the loop and repetitive labour.
Evmides et al. propose a framework which focuses
on processing big data in real-time [9]. It considers
collecting, processing, storing and analyzing AIS data.
Additionally, so-called intelligent services are utilized.
They cover, first, measurements based on descriptive
statistics, second, behaviors of vessels with respect to
spatially defined polygons and, lastly, collision
detection based on linear extrapolation vessels
motions.
In [10], [11] the authors propose a traffic
management priority index. This priority index aims at
indicating which situation requires focus most
urgently. In the [10], it is calculated by, first, clustering
vessels and, second, applying fuzzy logic to assess the
hazard level for given clusters, i.e., situations. The
priority index is grouped according to the hazard level
into attention, threat, danger or collision. In the work
of [11], the priority index represents a value between 0
and 1. It is calculated taking into account geometrical
relations such as closest point of approach (CPA), time
to closest point of approach (TCPA), encounter angles
and others.
Zhang and et al. propose a methodology which
consists of few steps [12]. Firstly, similarly to [10]
vessels are clustered. Secondly, for each cluster the
collision risk calculated by taking into account the
vessels attributes and Convention on the International
Regulations for Preventing Collisions at Sea
303
(COLREGs). Lastly, given on the calculated collision
risk a network model detects vessels with high levels
of risk. The authors state that this methodology is
adaptable to various port regions, i.e., territories.
The approaches in the related works show starting
points for increasing the situational awareness.
However, the approaches are highly specific and do
not consider multimodal processing. In contrast, our
demonstrator covers aspects which have been
proposed by the related works. Xiao et al. bring up
relevant requirements for a DST [6] which
independently have been identified and implemented
in project LEAS. This includes the incorporation of
ENC or creation of a knowledge bases. Evmides et al.
most importantly address the challenge of real- time
capability of a system which collects, processes, stores,
analyzes and, lastly, applies intelligent agents to a large
volume of (live) data [9]. However, the authors
primarily mention AIS data whereas the demonstrator
presented in this work considers further data sources.
In [10], [11] a priority index is proposed. Our
demonstrator sorts situation criticality according to
their hazard level through explicitly defined levels.
Moreover, our holistic approach in which a variety of
situations can be detected and combined, adapts
situations criticalities. This leads to a lower number of
false alerts.
3 PROJECT
The project goal is the development of an AI-based
DST for shore-based monitoring and steering of mixed
maritime traffic [13]. Mixed traffic refers to traffic
consisting of fully autonomous, semi-autonomous and
conventional participants. The demonstrator
incorporates the following goals and innovations:
− extensive support for VTSO through AI-based
methods
− adaptation of Human-Machine Interface (HMI)
with consideration of AI-based modules
− early detection of relevant hazardous situations
− consideration of mixed traffic
− creation of simulation possibilities for training of
(hazardous) situations with mixed traffic
The resulting demonstrator is evaluated by VTSOs
from German VTS centers Travemünde and
Warnemünde.
To achieve these goals project partners with
complementary expertise take part in the project which
are Fraunhofer CML, Fraunhofer FKIE, DLR-KN, DLR-
KI, HSW - University of Wismar and Bergmann Marine
as of today [13]. The project is completed in 40 months.
In Fig. 1 the work package (WP) structure of the project
is briefly illustrated. The work packages are executed
roughly in order from top to bottom as sequentially
depicted by the arrows. Only exception is WP7 which
covers legal research and is executed during the entire
project time.
At the time of preparation of this article, the
demonstrator is being evaluated. Thus the project is in
its final stages.
With respect to the project WP structure, the focus
of this paper lies on:
− WP2: Traffic Management Concept and System
Architecture, i.e., data fusion and preparation
− WP4: AI-based Systems and Modules for Decision
Support, i.e., anomaly detection and hazard
escalation
− WP5: Demonstrator for AI-based Modules for
Decision Support with innovative HMI, i.e.,
integration of AI modules into demonstrator and
demonstrator into VTS simulator
Figure 1. WP structure of the project LEAS. This paper focus
on the work packages illustrated by boxes with white
background.
The final results including the evaluation of the
demonstrator by VTSOs as test subjects will be
presented in a subsequent publication.
4 DEMONSTRATOR
In this chapter, the demonstrator is presented and
explained in detail. First, an overview of the system
architecture is given. The abstract system architecture
template and the actual implementation are explained.
The architecture plays a significant role concerning the
modularity, hence, extensibility to multimodal
operation of the demonstrator. Then, the module for
the situation detection and resolution is presented.
This is the core when it comes to the situation
awareness of the demonstrator. They underlying
formalism and its implementation are explained. Also,
an excerpt of currently detectable situations is given.
This chapter concludes with capabilities and
limitations of our approach.
4.1 System Architecture
The system architecture abstractly consists of
− data sources,
− the data fusion module,
− various AI modules and
− the HMI.
The connections of the composites are illustrated in
Fig. 2. The system architecture is designed accept
various data sources (illustrated in Fig. 2 with shaded
background). This enables the DST to work
multimodally and facilitates changes of the system
environment. With that approach it has been effortless
to switch demonstrator’s environments between a VTS
304
simulator or recorded traffic scenarios. To do so, a
central data fusion module processes incoming data
from various data sources, such as AIS data inside a
National Marine Electronics Association 0183 (NMEA)
data stream, radar data, VHF radio communication or
ENC, which typically are available at VTS centers (cf.
Tab. 1). The data fusion module works object- oriented
in that sense that each ship target represents an object.
It identifies tracks, and, if applicable, fuses ship targets.
The data fusion is able to fuse targets which are given
by AIS and radar data sources. Furthermore, the data
fusion allows for object-oriented annotation.
Inherently, it evaluates and annotates applicable rules
and actions according to COLREGs based on the
relative motion of any specified own ship and ship
target [14].
Figure 2. High-level demonstrator system architecture
abstractly illustrating interfaces.
The pre-processed data then can be requested by
any (AI) module as well as by the HMI. The connected
modules then can return their results by annotating the
given objects. Through this mechanism the modules
are able to exchange information and the HMI is able
to request the fused ship data which can be AIS and
radar data as well as the results from the connected
modules.
The presented demonstrator relies on that system
architecture. The specific implementation is illustrated
in Fig. 3. In the context of the project LEAS a VTS
simulator and an external microphone serve as data
sources. The VTS simulator delivers AIS data
encapsulated in NMEA sentences. Moreover,
recordings of the encapsulated AIS data can be
accepted by the data fusion. The data then is processed
by the data fusion which creates objects out of the ship
targets. Due to the data source agnostic approach
which relies on standards such as NMEA any VTS
which supports an NMEA interface.
Figure 3. Low-level demonstrator system architecture
depicting situation detection and resolution and its relevant
interfaces.
All depicted modules of the demonstrator run on a
Dell XPS notebook. It is equipped with 20 CPU cores
and 64 GB of RAM. The situation detection and
resolution module is running with a frequency within
an order of magnitude of one second. A significant
acceleration of the demonstrator can be expected when
the modules are optimized for performance which has
not been the focus so far.
The microphone can be used to imitate VHF radio
communication. A dedicated module transcribes the
spoken communication and then annotates it to any
ship target.
This paper focuses on the module for the situation
detection and resolution which is presented in detail in
the following chapter.
4.2 Situation Detection and Resolution
The AI module Situation Detection and Resolution (see
Fig. 3) relies on the formalism which has been
introduced by Stach et al. in the previous paper [5]. In
the following it is briefly introduced. However, for
more details, readers are referred to the previous
paper.
The formalism consists of three hierarchical spaces:
the object space 𝕆, the measurement space 𝕄 and the
situation space 𝕊. In Fig. 4 these hierarchical spaces,
their representations and interrelations are illustrated.
Figure 4. Illustration of formalism based on how the situation
detection is implemented.
The object space 𝕆 contains objects 𝑜
𝑖
which are
participants, i.e., vessels, or actors, e.g. physical or
regulatory obstacles, in maritime traffic. Objects have
attributes which can be determined directly or
indirectly. The determination of the attributes is called
measurement mi within the formalism. For example,
position is a measured quantity of a vessel and the
directly determined measurement value are the
latitude and longitude values. By contrast, the CPA
value of a pair of vessels is an indirect measurement
since it is based on other measurements such as
position, Speed over Ground (SoG) and Course over
Ground (CoG). The situation space 𝕆 contains
situations 𝑠
𝑖
which are assessments of measurements
or logical combinations of existing situations. A
specific situation is introduced when a measurement
fulfils the situations criteria. This mechanism is
illustrated in Fig. 5 left.
305
Figure 5. Hierarchical diagrams according to the used
formalism for traffic situations Noticeable CoG (left) and
Entering Offshore Wind Farm (OK) (right) based on the
introduced formalism.
Here, position and CoG of a vessel are measured.
Based on its position, the vessel is assigned to a
statistically parameterized grid cell. The CoG is taken
into account to verify if it matches the defined intervals
of the grid cell. If the CoG of the vessel does match the
defined intervals, then the situation Noticeable CoG is
introduced.
Situation criteria can also be based on the logic
combination of multiple situations instead of the
assessment of measurements. This facilitates intuitive
combination of situations, introduction of situations
with broad contextual consideration and reduction of
too generic situations. This mechanism is illustrated in
Fig. 5 right. There, a vessel and an ENC object are taken
into account. For the sake of simplicity, the
measurement space is skipped in the illustration. In the
situation space, initially the situations Noticeable CoG,
Is Operation & Maintenance Vessel and Entering Offshore
Wind Farm are introduced. In other words, given the
vessel position an unusual course is detected (since it
seems to be leaving a traffic separation scheme (TSS)
lane). However, the vessel is an operation &
maintenance vessel and its course is oriented towards
an offshore wind farm. Combining these pieces of
information it becomes apparent that the operation &
maintenance vessel is on its way to its designated wind
farm. Given that context, the vessels course is usual
and the situation Noticeable CoG can be dissolved.
Lastly, it should be noted that this formalism
inherently captures the causes for any situation in a
transparent manner. This facilitates the determination
of a resolution of the situation which is achieved by no
longer satisfying situation criteria. Following the
illustrations in Fig. 5, the causes and potential starting
points of resolutions are determined by going through
the diagrams from top to bottom.
In the presented demonstrator this formalism is
implemented based on a hybrid architecture: an expert
system with rules which utilize ML techniques. The
expert system is implemented using the Python library
experta [15]. Various subclasses of objects,
measurements and situations are introduced. In each
cycle of the expert system, situations or either
initialized or dissolved depending on whether their
criteria are fulfilled or no longer satisfied.
Each situation is assigned a criticality level
following the suggestion by the IALA which are with
ascending order of hazard level [4]:
1. caution
2. warning
3. alarm
4. emergency alarm
The specification of the hazard level of each
situation is defined considering insights from VTSO
and ergonomical approaches. For example, to reduce
the false or distractive alerts, the situation Entering
Offshore Wind Farm has the hazard level warning
whereas the context-enriched situation Entering
Offshore Wind Farm (OK) has the reduced hazard level
caution (cf. Fig. 5). However, the hazard levels can be
adapted easily if needed.
In the following, an excerpt of types of detectable
situations is given.
4.2.1 COLREGs
Based on AIS data COLREGs are assessed for each
vessel- vessel combination as proposed by Constapel et
al. in [14]. As a result, the situation detection and
resolution module annotates each vessel with the tags
crossed, crossing, giveway, head-on, overtaken,
overtaking, safe or standon with regards to any other
vessel, if applicable. The identified COLREGs
situations can be incorporated into the situation
resolution. Further details on the processing of
COLREGs assessments are explained in [14].
4.2.2 Area-related
In our implementation, the situation detection and
resolution module is directly connected to a ENC
server (see Fig. 3). This enables the extraction of ENC
objects. In accordance with the projects evaluation
traffic scenarios, point objects like AtoNs or area
objects such as military practice areas, offshore wind
farms, anchorage areas, TSS lanes or separation zones
and depth contours are extracted and taken into
account. The position and course of any vessel then is
put into relation with the extracted ENC objects. Thus
an area approach or penetration can be detected and
considered for further context information (cf. Fig. 5).
4.2.3 SoG- or CoG-related
The SoG and CoG of each vessel is assessed in
relation to empirically (manually) defined or
statistically determined criteria. Empirical criteria can
refer to either speed rules or minimum speed required
for manoeuvrability. Statistically determined criteria
make use of historical data. For the presented
demonstrator, AIS data has been indexed following a
hexagonal grid with the Python library H3 [16] and
then the distributions of CoG and SoG have been
approximated by Gaussian mixture models [17]. Based
on these Gaussian approximations confidence intervals
have been determined according to which the CoG or
SoG of a vessel is considered either normal or
anomalous, i.e., noticeable.
4.2.4 Collision Risk
A collision risk situation is detected based on CPA
and TCPA measurements. Analogous to SoG or CoG-
related situations, CPA and TCPA values are assessed
in relation to empirically (manually) defined or
statistically determined criteria.
The measurements of CPA and TCPA are applied
not only on pairs of vessels but also between vessels
and objects other than vessels, e.g., aid to navigations
(AtoNs) which are automatically extracted from ENC.
306
4.2.5 VHF
Spoken radio communication, i.e., over VHF radio,
can be transcribed applying the automatic speech
recognition system OpenAI Whisper [18]. Then
contextual information can be extracted from the
transcribed communication. The demonstrator covers
basic, however, common and relevant cases such as
green-to-green agreements or mayday calls.
4.3 Capabilities and Limitations
4.3.1 Agnostic towards Data Source
The presented demonstrator offers many
capabilities which have not been covered yet by other
works (cf. Chapter 2). From VTS center to VTS there
can be various communication and sensor technologies
as data sources available (cf. Tab. 1). Beyond the usage
of a standardized AIS or radar data interface, the
system architecture of the demonstrator facilitates the
integration of further data sources. This ensures that
the demonstrator is applicable at most VTS centers
independently from their available data source. The
integration of other data sources than AIS or radar
increases the level of information on the monitored
maritime traffic. This in turn increases the accuracy of
the situation detection and resolution so that false
alerts are reduced. The integration of VHF radio
communication as data source and its automatized
processing in a DST is a novel functionality.
4.3.2 Transparency and Interpretability
The underlying formalism of the situation detection
and resolution module as well as its implementation
into a hybrid expert system offers many advantages.
First and foremost, it makes the functioning of the DST
inherently transparent and its output interpretable.
This is a crucial feature because it makes the VTSO the
functioning and output of the DST understandable
This increases trust and acceptance into the DST. In
case of doubt, unclear or apparently wrong decisions
can be traced. The modular design of the situation
detection and resolution module also allows for easy
adaptions and extensions. For example, situation
criteria can be adapted or situations introduced resp.
removed whenever needed. Moreover, it facilitates
future validation and verification of the DST [19], [20].
4.3.3 Extensibility
Even though the situation detection and resolution
is an expert system, fundamentally, its rules can be
based on cutting-edge techniques such as ML. By this
hybrid approach the demonstrator remains
transparent and still enables the utilization of novel
and promising techniques. The presented
demonstrator utilizes ML-based techniques in the cases
of, first, SoG or CoG-related assessment and, second,
VHF transcription (see Ch. 4.2). In particular, the SoG
or CoG-related assessment based on automatic
statistical parameterization enables the application in
various VTS- territories with differing traffic patterns.
4.3.4 Intuitive Abstraction
Lastly, due to its hierarchical abstraction from
objects to measurements to situations the formalism
facilitates the intuitive combination of situations. This
particularly helpful as soon as territories with highly
specific or novel traffic patterns need to be monitored.
Based on our experience with the project, the situation
space enabled easy exchange about additions or
adaptions of detectable situations with project partners
and VTSO with less technical background.
4.3.5 Reliance on Standards
The demonstrator relies on standards such as
NMEA for AIS data streams. This may be seen as a
limitation but, however, facilitates the integration into
different VTS centers.
5 CONCLUSION & FUTURE WORK
All in all, the presented demonstrator delivers a more
holistic range of functionality and a higher Technology
Readiness Level (TRL) than related works (cf. Ch. 2). It
fulfils some of the requirements for a framework for a
next- generation VTS system as proposed by Xiao et al.
[6]. The demonstrator has been developed from scratch
which made it possible to design a system architecture
which aims at scalability and explainability. Thus,
given the compliance of common standards such as
NMEA data streams, the demonstrator can be
implemented at any VTS simulator or actual center.
Moreover, its situation detection and resolution
module is capable to adapt to territory-dependent
traffic patterns.
During the preparation of this paper, the usability
of the demonstrator is being evaluated following
evaluation processes described in DIN EN ISO 9241-
210. This is performed through interviews and
questionnaires with VTSOs during and after the
individual trials. Every trial consists of three traffic
scenarios which are approx. 20 minutes long and
include various hazardous traffic situations. Different
aspects of usability of the demonstrator are considered
which are, inter alia, task appropriateness, conformity
of expectations or transparency. The functionality of
the demonstrator has been evaluated during the
development and pretrials. It was tested and verified
that the demonstrator is able to detect every hazardous
situation (see situation types listed in Ch. 4.2.) in the
given traffic scenarios.
Their feedback will flow into future iterations of the
demonstrator. More complex traffic scenarios can be
evaluated using the European Maritime Simulator
Network (EMSN). Moreover, improvements will focus
on expanding its contextual understanding to support
a broader, more dynamic situation detection. The
objective is to move from isolated rule-based detections
to a comprehensive situation-context model that
enables real-time, multimodal awareness and allows
VTSO to interact with the system in a declarative,
human-centric manner.
5.1 Improvement of speech-to-text and text-to-context
One major direction of development is the
improvement of VHF speech-to-text and,
subsequently, text-to-context capabilities as introduced
in Chapter 4. This, on the one hand, refers to the
improvement of the speech-to-text functionality in
general. As addressed by Martius et al. this poses a
307
challenge since maritime radio communication data is
scarce [21]. The authors propose to enrich training data
by synthetic data. On the other hand, VHF radio
communication should not be treated as isolated
textual input. Rather it will serve as a fully integrated
component in the broader situation awareness model.
Utilizing methods as shown in [22], fusing VHF radio
communication with AIS will allow the system to
correlate further situations than the green-to-green
agreement or mayday call such as
− outputs from the situation and detection module, in
general,
− current port call schedules and berth allocations
and
− broadcast safety or navigational messages.
This can be fused into a coherent and queryable context
model which allows VTSOs to issue high-level queries
in natural language similar such as:
− "Highlight vessels with increased CPA risk that
have not yet responded on VHF radio."
− "Are there any critical situations near scheduled
arrivals within 30 minutes?"
− "List all vessels inside a restricted area without a
valid permission."
This approach aims to simplify the interaction
model for VTSO, enabling them to explore complex
situational combinations without the need to manually
correlate disparate data sources.
5.2 Use for Training Purposes
In addition to supporting operational decision making,
such a system has great potential as an embedded
learning tool. As VTS centers increasingly recruit
personnel directly from academic training into
operational roles, often without prior seafaring
experience due to the ongoing shortage of qualified
maritime professionals, this type of intelligent support
tool can help facilitate learning on the job. By making
situational context, navigational standards and
situation indicators transparent and accessible, it
enables inexperienced VTSOs to gain domain
knowledge organically during real- world surveillance
tasks.
ACKNOWLEDGEMENT
These research results are part of the project LEAS [13]. The
project is funded by the German Federal Ministry of
Education and Research (BMBF) within the programm
“Research for Civil Security 2018-2023” under grant number
13N16246 managed by VDI Technologiezentrum.
REFERENCES
[1] “Review of Maritime Transport 2024,” United Nations
Conference on Trade and Development, Geneva, Tech.
Rep., 2024.
[2] “VTS Manual,” International Association of Marine Aids
to Navigation and Lighthouse Authorities, Tech. Rep.,
2021.
[3] International Association of Marine Aids to Navigation
and Lighthouse Authorities, “G1111 Preparation of
Operational and Technical Performance Requirements
for VTS Systems,” International Association of Marine
Aids to Navigation and Lighthouse Authorities, Tech.
Rep., 2022.
[4] ——, “G1110 Use of Decision Support Tools for VTS
Personnel,” International Association of Marine Aids to
Navigation and Lighthouse Authorities, Tech. Rep., 2022.
[5] T. Stach, P. Koch, and M. Constapel, “A Formalism for
Scalable Maritime Traffic Monitoring and Explainable
Anomaly Detection and Resolution at Vessel Traffic
Services,” Journal of Physics: Conference Series, vol. 2867,
no. 1, p. 012053, 2024.
[6] Z. Xiao, X. Fu, L. Zhao, L. Zhang, T. K. Teo, N. Li, W.
Zhang, and Z. Qin, “Next-Generation Vessel Traffic
Services Systems—From “Passive” to “Proactive”,” IEEE
Intelligent Transportation Systems Magazine, vol. 15, no.
1, pp. 363–377, Jan. 2023.
[7] A. Kypriotaki, “BSU Investigation: Collision of vessel
with offshore wind turbine,” Jan. 2025.
[8] T. Stach, Y. Kinkel, M. Constapel, and H.-C. Burmeister,
“Maritime Anomaly Detection for Vessel Traffic Services:
A Survey,” Journal of Marine Science and Engineering,
vol. 11, no. 6, p. 1174, Jun. 2023.
[9] N. Evmides, L. Odysseos, M. P. Michaelides, and H.
Herodotou, “An Intelligent Framework for Vessel Traffic
Monitoring Using AIS Data,” in 2022 23rd IEEE
International Conference on Mobile Data Management
(MDM), Jun. 2022, pp. 413–418.
[10] L.-n. Lee and J.-s. Kim, “Development of Priority Index
for Intelligent Vessel Traffic Monitoring System in Vessel
Traffic Service Areas,” Applied Sciences, vol. 12, no. 8, p.
3807, Jan. 2022.
[11] L. Zhao, S. R. T. Sunilkumar, B. Wu, G. Li, and H. Zhang,
“Toward an Online Decision Support System to Improve
Collision Risk Assessment at Sea,” IEEE Intelligent
Transportation Systems Magazine, vol. 15, no. 2, pp. 137–
148, Mar. 2023.
[12] L. Zhang, J. Ge, F. Goerlandt, L. Du, T. Chen, T. Gu, L.
Gan, and X. Li, “A Method for Enhancing the Traffic
Situation Awareness of Vessel Traffic Service Operators
by Identifying High Risk Ships in Complex Navigation
Conditions,” Journal of Marine Science and Engineering,
vol. 13, no. 2, p. 379, Feb. 2025.
[13] “LEAS - Fraunhofer CML,”
https://www.cml.fraunhofer.de/en/research-
projects/LEAS.html.
[14] M. Constapel, P. Koch, and H.-C. Burmeister, “On the
implementation of a rule-based system to perform
assessment of COLREGs onboard maritime autonomous
surface ships,” Journal of Physics: Conference Series, vol.
2311, no. 1, p. 012033, Jul. 2022.
[15] R.A. M.Pérez, “Nilp0inter/experta,”
https://github.com/nilp0inter/experta, Apr. 2025.
[16] I. Brodsky, “H3: Uber’s Hexagonal Hierarchical Spatial
Index,” https://eng.uber.com/h3/, Jun. 2018.
[17] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B.
Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss,
V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M.
Brucher, M. Perrot, and E. Duchesnay, “Scikit-learn:
Machine Learning in Python,” Journal of Machine
Learning Research, vol. 12, no. 85, pp. 2825–2830, 2011.
[18] A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey,
and I. Sutskever, “Robust Speech Recognition via Large-
Scale Weak Supervision,” Dec. 2022.
[19] E. Commission, “Regulation (EU) 2024/1689 of the
European Parliament and of the Council of 13 June 2024
laying down harmonised rules on artificial intelligence
and amending Regulations (EC) No 300/2008, (EU) No
167/2013, (EU) No 168/2013, (EU) 2018/858, (EU)
2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU,
(EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence
Act) (Text with EEA relevance),” 2024.
[20] P. Koch, T. Stach, M. Constapel, and H.-C. Burmeister,
“Studie zur zielbasierten Standardisierung in der
Prüfung und Zulassung intelligenter
Entscheidungseinrichtungen von teilautonomen
Überwasserfahrzeugen,” Tech. Rep., 2023.
308
[21] C. Martius, E. C¸ . Nakilcioglu, M. Reimann, and O. John,
“Refining maritime Automatic Speech Recognition by
leveraging synthetic speech,” Maritime Transport
Research, vol. 7, p. 100114, Dec. 2024.
[22] Y. Chen, X. Qi, C. Huang, and J. Zheng, “A data fusion
method for maritime traffic surveillance: The fusion of
AIS data and VHF speech information,” Ocean
Engineering, vol. 311, p. 118953, Nov. 2024.