Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No.

2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

A Review in Automatic Visual Inspection for Railway Overhead Contact

Line Systems based on Image Processing

Achmad Romadoni1, Satrio Wicaksono2, F. Ferryanto2

Graduate School, Faculty of Mechanical and Aerospace Engineering, Institut Teknologi Bandung, Jalan
Ganesha 10, Bandung 41032, Indonesia1,
Mechanical Design Research Group, Faculty of Mechanical and Aerospace Engineering, Institut
Teknologi Bandung, Jalan Ganesha 10, Bandung 41032, Indonesia2
Corresponding Author: [email protected]

ABSTRAK
Inspeksi sistem overhead catenary meliputi pemeriksaan pada geometri kawat kontak, interaksi antara kawat kontak dan
pantograf, cacat pada komponen, komponen yang aus, dan jarak bebas sangat diperlukan untuk memastikan keandalan,
ketersediaan, pemeliharaan keselamatan infrastruktur dan operasi kereta api. Teknologi inspeksi visual otomatis pada sistem
overhead catenary dapat meningkatkan efisiensi, efektivitas biaya, dan presisi jika dibandingkan dengan metode inspeksi
konvensional. Makalah ini memberikan gambaran umum dan kontribusi penelitian yang dilakukan oleh para ahli di bidang ini,
serta aplikasi dan kemajuan teknologi inspeksi visual otomatis untuk sistem overhead catenary kereta api. Proyeksi arah
penelitian di masa depan untuk sistem inspeksi otomatis dalam kegiatan inspeksi sistem catenary overhead juga akan dibahas.
Kunci: Sistem Overhead Catenary, Inspeksi Kereta Api, Visi Komputer, Inspeksi Otomatis.

ABSTRACT
Inspections of overhead catenary systems include checks on the geometry of contact wire, the interaction between contact wire
and pantograph, defects in components, worn components, and clearance are necessary to ensure the reliability, availability,
maintainability, and safety of railways infrastructure and operation. An automatic visual inspection technology of overhead
catenary systems can improve conventional inspection methods' efficiency, cost-effectiveness, and precision. This paper provides
an overview and contributions of the research made by scholars in this field, as well as the application and advancement of
automated visual inspection technology for railway overhead catenary systems. The projection of the future research direction
for automatic inspection in overhead catenary system inspection activities was also provided.
Keywords: Overhead Catenary Systems; Railway Inspection; Computer Vision; Automatic Inspection.

one way to maintain OCS reliability, but their

1 INTRODUCTION availability will decrease due to frequent inspection
activities [6]. Moreover, if inspections are
Overhead Contact Line System (OCS) transmits
conventionally performed, complications may arise
electric power from the traction substation to electric
due to the massive number of assets that should be
trains as the primary energy. Various types of OCS are
inspected, driven maintenance costs, and some
implemented according to the operating system and its
inspections may endanger workers. Some railway
purpose, such as simple catenary, stitched catenary,
companies perform condition-based maintenance to
compound catenary, and rigid catenary or conductor
maintain OCS performance as one of the solutions
rail [1][2][3]. The growth of OCS networks worldwide
[7][8]. Furthermore, predictive maintenance can be
is increasing rapidly. Their reliability, availability,
performed after consistently collecting historical data
maintainability, and safety must be maintained
from many inspection activities [9][10][11].
because degradation of the existing OCS network
Therefore, an automatic inspection method based on
equipment due to continuous operations and exposure
computer vision is needed to enhance the maintenance
to environmental conditions will reduce performance
methods mentioned.
over time [4][5]. Increasing inspection frequency is

50
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

Computer vision is a technology that can be utilized and state of the OCS line ahead. Various intelligent
for non-contact measurement. Several studies on image recognition analyses will process the acquired
railway inspection using computer vision have been image or video data to obtain the needed parameters
widely developed. Some authors have reviewed the and utilize some algorithms to identify abnormalities.
use of computer vision for railway inspection Some methods have even pinpointed abnormal
purposes. Liu et al. summarized the use of image locations and automatically generated reports.
processing technology for detecting faults and defects Generally, to support the computer vision-based
in railway tracks, pantographs, catenary networks, automatic inspection process, it uses equipment such
train body parts, and infrastructure inspections such as as cameras, illumination, optical devices: filters and
stations, crossings, and tunnels [12]. Besides that, Gao lenses, processing computers, communication
et al. provide a general overview of the automatic interfaces, and mounting equipment. The differences
detection and monitoring system for the pantograph in the specifications of the equipment installed must
and catenary system in China's high-speed railways. be adapted to meet the inspection requirement and the
There are six types of inspection equipment and their object to be inspected, including camera orientation,
applications, which are: (1C) comprehensive angles, and even image processing methods to achieve
pantograph and catenary monitoring, (2C) catenary the best results and enable all weather conditions.
checking video monitoring, (3C) catenary checking Some studies on OCS inspection using computer
online monitoring, (4C) high precision catenary vision may have pros and cons results that still need to
checking monitoring, (5C) pantograph and catenary be discussed, such as measurement accuracy,
video monitoring, and (6C) ground monitoring for precision, robust detection method, and fast
catenary and power supply equipment. This 6C system computing processes for online real-time control
can identify potential defects and assess the conditions that require improvement.
pantograph and catenary system's operational efficacy
to ensure safe operation [13][14]. Study on the details specific to OCS inspection is still
very limited. Thus, several methods regarding
Danijela et al. present a review analysis of obstacle automatic visual inspection for railways OCS
detection and distance estimation techniques for especially based on computer vision will be discussed
railways. This discussion aims to reduce the number in detail. This study aims to provide information on
of collisions between trains and particular objects. The the development of OCS inspection, then insight into
explanation focuses on the currently used sensors, the most viable and frequently used methods to obtain
particularly vision sensors, and categorizes the good results depending on the object inspected and the
methods into traditional Computer Vision-based and software and hardware implemented so far.
Artificial Intelligence-based methods. Each
classification entails three key aspects: (1) Rail track 2 RAILWAY OCS AND ITS INSPECTION
extraction, (2) Identifying obstacles on or near rail
tracks, and (3) Estimating distances between onboard 2.1 Object of OCS inspection
cameras and identified obstacles [15]. Although the
The OCS determines power traction supply
discussion presented by the author is still about
performance. There are three main objects: OCS,
obstacles in the track area, there are still possibilities
pantograph, and the interaction between OCS and
to be applied to the OCS system. Unlike before, Long
pantograph [5]. OCS inspections are normally
et al. analyzed existing computer vision techniques
performed to check: geometry parameters, supporting
focusing on two critical areas: measuring overhead
equipment, insulator, contact wire wear, wind
contact wire parameters and identifying overhead
deviation, abnormal fitting, foreign object, and the
contact wire conditions. The overhead contact wire
interaction between OCS and pantograph (contact
geometry parameters can be precisely measured using
force, arch, and collector strips) [16]. The current
methods based on stereo vision, except in particularly
collection process requires maintaining the geometric
extreme conditions such as train motion caused by the
parameters of the contact wire within specified limits.
sway, vibration, rotational angle on train body
The geometry parameter constraint ensures that the
dynamics, and environment variation or complex
contact point between the contact wire and the current
background [4]. Then the next chapter will discuss in
collector remains in its proper position. So, this is the
more detail the research progress in OCS inspection
reason why the inspection of several aspects above is
on railways.
imperative.
A common setup camera is mounted on the rail vehicle
The initial inspection object is the contact wire's
roof or driver's cabin train to capture the object of
geometry parameter (CGP). CGP is generally divided
inspection, such as the pantograph, OCS component,
into stagger and height [17], creating risks if poorly
51
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

maintained. The stagger of the contact wire refers to

the lateral displacement of the wire point of contact
from the rail track center. The stagger of contact wire
is set up in a zigzag shape to prevent local wear on the
current collector strip pantograph. The contact wire's
stagger is also maintained to avoid exceeding lateral
displacement from the permissible limits that can
cause the contact wire not to be in its proper position.
If the contact wire has excess stagger, it may lead to
poor current conductivity, cause a spark over or arcs,
and even break the contact wire. Besides, it is essential Figure 1 Contact wire’s height and pantograph interactions.
to maintain the contact wire's height to prevent contact
loss. The contact wire's height refers to the vertical
distance from the rail track's top surface to the contact
wire's lower surface while measured perpendicularly.
The contact loss of contact wire also may lead to poor
conductivity. Conversely, if the contact wire's height
is too low, it may lead to intense friction on the contact
point and accelerate wear, causing arcs and even the
pantograph to stop operating. An illustration of the
CGP is shown in Figure 1 - Contact wire’s height and
pantograph interactions and Figure 2 -Contact wire's
stagger and contact point. The typical nominal CGP Figure 2 Contact wire's stagger and contact point [19][20]
parameter for simple catenary is shown in Table 1.
Table 1 Nominal CGP parameter on simple type OCS [21]
The second inspection object is the OCS supporting
equipment, which acts as a support structure that helps Simple Type Catenary
Nominal Unit
maintain the performance of the OCS. OCS support Geometry Parameter
equipment includes various components such as Span length 50-60 m
insulators, pins, bolts, clamps, and tension fittings. Contact wire stagger
200 mm
The essential activity of this inspection is to recognize (straight)
Contact wire stagger
the type of component and its condition, for example, 300 mm
(Curve)
loosening or missing pins and bolts, broken clamp and Contact wire height 4.3 – 5.7 m
tension fitting, fracture, or corroded insulator, Max. gradient 6 ‰
deflection of arm and other supporting structure. This
Max. change in gradient 3 ‰
type of inspection is also often implemented to assess
new installations.

The next inspection object is the sliding condition Midya et al. explained a comprehensive study of arcs
between the pantograph and the contact wire. and their mechanisms. They do laboratory-scale
Specifically, observe the interactions on the current experiments to test many parameters that may cause
collector strip of the pantograph and the contact wire arcs on the pantograph [22]. As well as the vertical
when the train is running. Some phenomena that will force of the pantograph called the pantograph's contact
occur during sliding are arcs, hot spots, and contact force. The way to determine the presence of contact
force. An electric arc is a current flow through an air force is by measuring the sudden acceleration of
gap between a contact strip and a contact wire, usually vertical displacement on the pantograph, or it can also
indicated by intense light emission. Arc occurs due to be by measuring the contact wire supporting structure
poor geometric adjustment and excessive friction, uplift. Using a finite element model, Bai et al.
which was previously briefly explained, sometimes analyzed the effect of contact wire design parameters
caused by improper static contact force, incorrectly set such as contact wire span length, height, and tension
aerodynamics, and worn components [18]. on contact force at various speeds. The results show
that the longer the span, the higher the contact force,
and the lower the trolley wire tension, the higher the
contact force. However, the difference in structure
height has no significant effect on contact force [23].

52
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

The conventional detection approach to identifying

issues with pantograph and catenary relies on manual
work. The inspector walks and utilizes insulated poles
or so-called Hot-Sticks to measure the stagger and the
height of the contact wire [17]. At the same time, the
inspector checks the environmental conditions around
the OCS. Some advantages of manual detection are
that it offers greater versatility and enables the
identification of diverse failure modes, but it suffers
from low efficiency and inadequate security.
Moreover, the long duration of conventional
inspection shall be scheduled to not coincide with train
operation time. Furthermore, subjective human factors
can impact detection outcomes, such as the experience
of the personnel involved.

Figure 3 Rollingstock clearance-narrow gauge [24]

The last inspection object is OCS's clearance and

intrusion. OCS's clearance is the safe distance between
live parts of contact lines and the supporting structures Figure 4 Hierarchy of OCS inspection methods.
or rail vehicles. Sometimes the building structures
The contact detection method on inspection is to check
from property development and some vegetation or
the kinds of parameters using measurement equipment
tree branches around the railway track could interfere
in direct contact with the pantograph or catenary. The
with train operations. Figure 3 describes the common
contact detection method on inspection is to check the
clearance between vehicle and the environment.
types of parameters using measurement equipment in
direct contact with the pantograph or catenary. The
2.2 Development of OCS Inspection contact detection method evolved with the
Utilizing measurement devices is integral to the advancement of railway electrification. Switch
pantograph and catenary inspection activities. Some sensors mounted on the pantograph can detect stagger.
traditional detection, contact detection, non-contact For example, when the contact wire has excess
detection, and combined detection methods already stagger, the contact wire passes through the adjacent
provide many advantages. Conducting efficient, switch sensor. Then it emits a signal that can be used
precise, real-time detection and monitoring to acquire as information to calculate and measure the stagger
accurate information regarding OCS operation is value. Like stagger detection, a displacement sensor
crucial to ensuring system safety. The resulting data can detect the contact wire's height by detecting the tilt
can be used to perform maintenance and address any change. The variation in contact wire height will cause
unfavorable conditions promptly. The development of a difference in the angle of the pantograph's arm and
pantograph and catenary detection has followed a the horizontal plane, which can help calculate and
progression of four different phases, as shown in measure the height value [16].
Figure 4. Jing et al. have reviewed the various uses of
inspection robots in the railway field. The sensors used Abdullah et al. conducted a laboratory experiment by
are various, such as Computer vision, Digital Image creating a scale model to identify the performance of
Correlation, Ultrasonic, Laser-vision System and laser the dynamic interaction of contact wire and
distance, Light Detection and Ranging (LiDAR), pantograph using load sensors and displacement
Accelerometer, thermal sensor, Eddy Current, sensors that combined with a laser to detect contact
Electromagnetic Acoustic Transducer (EMAT), and force. The data collected is used to trigger the actuator
soon [25]. as active control of the pantograph [26]. Donnell and
Sing et al. have developed an inspection system called
PANDAS (Pantograph damage assessment system).

53
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

This module system is integrated into the pantograph examination of OCS. The inspected objects depend on
to detect and monitor wear and damage that occurs the part of OCS that needs to be periodically checked.
[27]. Besides that, Marco et al. implement optical fiber The image data acquired from the inspection will be
sensors for strain measurement, enabling real-time processed in real-time or sent to the office to be
monitoring to detect contact force. There are two types analyzed offline. The automatic visual inspection of
of fiber sensors, which are fiber brag gratings sensors OCS requires stages such as (1) data acquisition by
(FBGs) and interferometric schemes[28][29]. Sung et taking pictures of an object using a camera or video,
al. conduct a study using a load cell and accelerometer (2) pre-processing by refining data or images
to measure contact force on a pantograph by adjustment, (3) segmentation by selecting the focused
considering the aerodynamic lifting force [30]. The object, (4) feature extraction to take reference of the
inspection system in the contact detection method is image for image matching, (4) object detection and
only able to detect one object. This method has some object recognition by training or labeling the input
disadvantages in durability and operation safety due to data, (5) processing data to produce good qualitative
poor temperature characteristics and unstable or quantitative output as desired [33]. Researchers use
connectors of the sensors, so it is feared that it can specific methods to produce the desired output under
interfere with the measurement results, and additional different parameters and environments at each stage.
devices must be used if the inspection is needed for For example, Ge et al. propose a computer vision
different objects. Therefore, non-contact methods system encompassing a modified version of faster R-
were introduced. CNN specific for the subway in real-time to inspect
pantograph interaction. Their research findings
Non-contact detection methods using ranging indicate that their approach has achieved impressive
technology such as laser, infrared sensor, or camera results, with about 94.9% detection precision [34].
and ultrasonic are kept advancing. Many non-contact Successful image detection of OCS components such
detection methods have become widely employed for as pantographs is important for further inspection
pantograph and catenary system detection. Peng et al. tools-based computer vision development. The next
implement an optical system that employs the laser subsequent segment will discuss how researchers
phase ranging technique to inspect contact wire utilize computer vision technology on various objects
gradients. This system will be installed on the low- on OCS inspection. Some recent research reviews
voltage side of the locomotive top and measure the have even used a combination of non-contact
distance target on the pantograph [31]. Ötslund et al. detection methods, both with audio and video as input,
propose a technique for condition-based maintenance and use Deep Learning to detect defects in the field of
of the pantograph current collector strip employing railway maintenance [35][36].
photodiodes. By establishing a relationship between
the intensity of arc light, DC components, and the 3 METHOD AND REVIEW OF RAILWAY
running distance, the wear of the current collector strip OCS INSPECTION BASED ON COMPUTER
of the pantograph is well predicted and increases VISION
availability more than the previous method, which is
limited to mileage intervals only [6]. This review utilizes a narrative approach to present a
synthesis of the latest knowledge related to automatic
Puschman et al. proposed the technology of measuring visual inspection of railway overhead contact systems
the lateral position of the trolley wire using ultrasonic based on image processing. The research integrates
and dedicated software called OVHWizard and FMA. various relevant literature to provide a comprehensive
This tool has a high accuracy compared to a laser, overview of the technological advancements in this
depending on the speed during measurement, but its field. The literature selection process involved a
weakness is that it cannot be further than 50 cm from thorough search across multiple academic databases
the contact wire [32]. Non-contact measurement using using relevant keywords, and the chosen articles were
ranging technology provides many advantages and evaluated based on their contribution and relevance to
guarantees detection activities with high efficiency the discussed topic. Through this approach, we aim to
and little traffic disruption. However, the simple offer a comprehensive understanding of how image
features and lack of accuracy of the sensors, as well as processing technology has been used to enhance the
the complicated data processing, require more efficiency and accuracy of railway overhead contact
working time for inspectors. Therefore, non-contact line inspections.
methods using camera technology were developed.
The general approach to computer vision is to make
The latest is a non-contact detection method using the computer equipment recognize an object and
image processing. Image processing technology has a perform specific tasks according to the user's purpose.
remarkable impact on the automated visual Figure 5 shows the components of the computer

54
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

vision system are a sensing device and interpreting Figure 8 shows the installation of an inspection
device [37]. Figure 6 shows how computer vision on detector on the roof of a rail vehicle. Not all types of
railway OCS inspection conducted. Figure 7 inspections in the railway OCS can be replaced using
illustrates clearance detection using image processing computer vision, such as detailed geometry
that showing another method that applied on Railway measurements on cross-over equipment and power
OCS inspection. The importance of conducting quality performance tests. However, most visual and
inspections is the standard as a reference to avoid the measurement activities can be replaced by computer
subjectivity of the inspector, so the several types of vision. The following sub-sections will review some
inspection discussed earlier are regulated in BS EN measurement parameters that can be performed using
50317. computer vision.

3.1 Geometry Parameter Inspection

Aydin et al. successfully implemented some methods
for monitoring anomalies on the pantograph and
contact wire. They combined some methods, such as
mean-shift for object tracking on contact wire [19]. In
Figure 5 Diagram of how computer vision generally works. other research, they also apply a modified kernel-
based method, blob analysis to detect arcs, and S-
transform to detect irregularity of contact wire
simultaneously. Gaussian mixture models (GMMs)
are also used to detect the foreground in each frame
[40]. Cho et al. propose a novel technique for the
Contact wire & Camera Computer Report
Pantograph dynamic stagger of contact wires utilizing a video
monitoring methodology. This approach addresses the
Figure 6 Diagram of how computer vision measures stagger on limitations of traditional methods, which may result in
contact wire significant inaccuracies when applied to tilting trains.
They are adding feature-based image matching
techniques such as the scale-invariant feature
transform (SIFT) algorithm and gradient local affine
invariant feature extraction (G-LAIFE) that are
resilient to changes in camera viewpoint. The
experimental video data was obtained from a camera
placed on the roof of the Korean Tilting Train Express
(TTX) and linked to the monitoring center within the
train. Processing image data is currently employed
offline, showing a correct detection rate of
approximately 85% [41].

Stela et al. proposed an intelligent software

application that can accurately determine the status of
contact wires and perform maintenance work based on
system-generated indications. The zigzag pattern was
determined using a camera to capture geometry
Figure 7 Illustration of clearance inspection [38]
measurements and conducted using the NI MyRIO
acquisition board [42]. The intelligent software
application will analyze the measured data and
provide two types of decisions: immediate
interventions at locations where defects are detected,
based on the most recent measurement, and
predictions regarding the future development of
contact wire conditions [43]. Liu et al. have developed
a high-precision detection of catenary geometry
parameters based on computer vision for electric
railways. They utilize laser-emitted spotted images to
Figure 8 Arcs Detector location on roof of vehicle BS EN 50317 determine the coordinates of the wire. The tracking
[39]
and locating spotted image method used is particle
55
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

swarm optimization - Generic particle filter algorithm inspection time conditions. Some defects were
(PSO-GPAF). Then added, the Kalman filter (KF) deliberately pre-made, and a total image was taken as
method was used as a correction factor from the many as 21024 images or 87.1 kilometers equivalent.
geometry parameter values in the form of stagger and Eighteen high-resolution cameras were installed on a
contact wire height that has been defined. The catenary ladder in various viewing directions during
resulting judgment of root mean square error (RMSE) the day and night conditions. Of course, the data can
after correcting for contact wire height is 0.7 mm, and be used as training for automatic visual inspection
for stagger is 1.3 mm [20]. As previously, Zhan et al. when the OCS is already operating in the future [47].
used a binocular vision camera (BVC) integrated with Liu et al. have introduced a method to detect fastener
a line scan camera (LSC). The method used is based defects on catenary support devices based on
on a 1-D target combined with a 1-D displacement computer vision. The method consists of three stages:
platform and adding a vibration compensation method (1) localize the joint of catenary supporting structures
(MDOF) to reduce the vibration error of the vehicle. using single shot multi-box detector (SSD)
The resulting judgment of root mean square error framework, (2) localize the fasteners using You Only
(RMSE) for contact wire height is 1.5 mm, and for Look Once (YOLO) as a classifier, (3) diagnose a
stagger is 2.20 mm [44]. defect of fasteners using deep convolutional neural
networks (DCNNs). Some fasteners, such as two
Huang et al. proposed a novel method for detecting screws, puller bolts, α-pin, β-pin, and nut, were
contact points (CPD) between the pantograph and successfully detected. The experiment data was taken
catenary with complex infrared images with high by XLN4C-01 imaging inspection vehicle during the
accuracy. The method comprises three primary night, and the proposed DCNNs have achieved an
elements: (1) separating single input called horizontal- impressive accuracy of 92.78 mAP. This method was
vertical enhancement (HVE) approach to be two also successfully applied for contact point detection
model outputs, (2) the vertical output model for point [48][49]. Gao et al. present a novel system for
detection of the pantograph, and (3) the horizontal detecting defects on insulator surfaces. This basic
output model for point detection of the contact wire. system employs some methods such as faster R-CNN
Then both components were detected using improved to localize the main component containing the
random sample consensus (RANSAC) to obtain their insulator, supervised deep material classifier (DMC),
contact points. All infrared data for this study were and unsupervised deep denoising autoencoder
obtained from the CRH2A electric multiple unit (DDAE) to classifier the score of anomalies. The data
(EMU) at the Nanchang Passenger Transport Section. set used in this study consists of approximately 18,000
The proposed scheme is efficient and robust, with an images taken by KCIS-01 catenary inspection vehicle
average pixel error of 0.48 and an average accuracy of during the night and processed offline. The
99.6% in two datasets comprising 12,000 frames. experiment features TensorFlow deep learning
Explicit recognition of the point of contact plays a framework, and the results achieved an F1 score of
significant role in monitoring the operation of EMU 0.95 [50]. Ding et al. propose a catenary defect
trains [45]. Zang et al. present a robust method for detection algorithm for high-speed railways. The
monitoring the condition of pantograph and catenary method consists of pixel region cutting and tilt
interaction based on a combined deep convolutional correction as pre-processing image, faster and mask R-
network with modified features to enhance the CNN to localized object detection, and using gradient,
stability of contact point detection in the presence of texture, and grey feature fusion (GTGFF) and K-
complex backgrounds, such as messenger wire, portal means clustering analysis model to detect the defects
structure, vegetation, tunnel environments with cloud, such as broken insulators, foreign bodies, dirt, and
glare, and other complex backgrounds. The modified flashovers. The insulator defect detection algorithm
features are the deep pantograph detection network was evaluated using 1500 isolator image datasets
(DPDN) and the deep pantograph segmentation taken by an inspection vehicle and demonstrated
network (DPSN). The proposed method achieved exceptional results, achieving a Recall of 99.0% and a
outstanding results with a segmentation performance Precision of 92.5% [51].
of about 90.2% mIoU and 5.5% error detection [46].
3.3 Contact Force and Arcs Detection
3.2 OCS Component Defect Detection
The interaction between the pantograph and the
The number of catenary component image databases contact wire in the OCS is inseparable. When an arc
is important for training machine learning. Wang et al. occurs, excess or lack of contact force can be one of
completed a database of catenary components for the causes. Landi et al. utilized a thermo-camera to
visual inspection during construction encompassing facilitate the monitoring of pantographs from
diverse catenary components, defect types, and overheating and arcs. They also combine cany edge,
56
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

and Hough transform methods to detect irregularity of train. An algorithm based on Helmholtz shears is
contact wires, as discussed before [52][53]. Koyama employed to detect contact wire. In order to acquire
et al. present a novel approach to quantify contact the operating speed of the train within 120-330 km/h,
force using a line sensor camera (LCS). The the transfer rate data requires 30 Mbytes / sec. This
pantograph panhead is marked by black strips to notice proposed method results in 80% corrected detection of
the vertical displacement. The vertical displacement an obstacle [59]. The illustration of pantograph
can be determined with normalized cross-correlation clearance intrusion is shown in Error! Reference
(NCC) as pattern matching. Additional equipment, source not found.. Wu et al. proposed a method to
such as load cells and illumination, is also necessary detect bird nests. They employed a novel framework
to verify this method. The experimental findings as follows: (1) adaptive binarization, (2) trunk/branch
showed good results, with a measurement accuracy of detection, (3) hovering point detection, (4) streak
91.5% [54]. Aydin et al. presented a method to extraction, and (5) pattern learning. They also utilized
maintain contact force between the catenary system the histogram of the orientation of streaks (HOS) to
and the pantograph by managing the height of the capture the distributions of orientations and the
pantograph. They applied the canny edge detection histogram length of streaks (HLS) to the distances of
algorithm and Hough transform to detect the height of detected twig streaks. The patterns of bird nests are
the running pantograph. The experiment was learned with SVM. The proposed method has been
processed in MATLAB Simulink and was suitable to evaluated, and the results have shown promising
simulate the vertical movement of the active control of performance with a precision of 35.82% and a recall
the pantograph in real-time operation requirements of 91.06%. The intrusion of bird nests is illustrated in
[55]. Another way that Aydin et al. do. They propose Figure 10 [60]. Tian et al. proposed a method of object
methods to detect pantographs and arcs using a firefly detection in the railway environment that aims to
optimization algorithm, integral image, and Haar-like improve the safety of railway operations. The method
feature as fitness detection functions. Based on the used is Variable focus multi-scale augmentation
findings, the firefly-based method has achieved a (VCMF) with functions to detect objects at various
satisfactory detection performance, with an accuracy scales. Then the feature extraction and fusion neural
rate of around 96%, which is suitable for practical network architecture using CentreNet (FFCN) serves
purposes. The method is observed in the laboratory but to improve object detection abilities. This method was
could be effectively implemented in real-time tested using the PASCAL VOC2007 database and
scenarios [56]. Arthington et al. utilized an onboard obtained an 81% mAP score[61].
camera to detect contact force. The method works by
detecting two markers mounted on the outer side of the
pantograph that travels up and down according to the
trajectory. Dynamic pantograph modeling is required
to perform force estimation, and the results are
optimized again using a Kalman filter to reduce noise.
The resulting output is quite good, approaching the
actual force [57]. Karakose et al. presented a novel
Figure 9 OCS’s obstacle detection [59].
approach employing Deep Learning based on
Convolutional Neural Network (CNN) for detecting
arc. They applied the five pooling layers architecture
of CNN in this method. Remarkably the experimental
findings demonstrate significant accuracy results, with
97% of the arc area correctly identified [58].

3.4 Clearance and Intrusion Detection

Moller et al. describe a method to detect catenary
obstacles using a tri-ocular camera sensor system. The
catenary obstacle detected consists of wooden knots,
plastic film, crime acts, and defective components
such as down hanging, ripped-off droppers, bonding, Figure 10 Bird nests detection on OCS [60].
and broken insulator. This method aims to retract the
pantograph immediately after an obstacle is detected 4 DISCUSSION
to avoid pantograph damage. Three progressive scan Based on several articles we reviewed, we have
cameras record the scene with zooming lenses to compared the advantages and disadvantages of
capture the scene about eighty meters ahead of the
57
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

conventional methods versus camera-based methods

as shown in the Table 2.

Table 2 Comparation of Advantages and Disadvantages between Conventional Inspection

Aspect Conventional Method Computer Vision Method

Advantages Simplicity: Straightforward Efficiency: Quick inspection of large areas
implementation [25]
Cost-Effective: Lower initial investment Consistency: Standardized inspection criteria [25]
[25]
Reliability: Established procedures and Detail and Precision: High-resolution imaging for
protocols detailed analysis [62]
Human Judgment: Nuanced decision- Safety: Reduces human exposure to hazards [63]
making based on experience [64] [25]
Disadvantages Time-Consuming: Slow and labor- Initial Cost: Investment in high-quality cameras
intensive [65][64] and systems [60][8]
Subjectivity: Variable results depending Technical Expertise: Specialized knowledge
on inspector’s skill and fatigue [65] required [66]
Limited Accessibility: Difficult to inspect Data Management: Needs robust storage and
hard-to-reach areas [60] processing capabilities [62][67]
Safety Risks: Potential hazards for Initial Setup: Time-consuming calibration and
inspectors in dangerous environments setup [62] [68]

5 CONCLUSIONS will evolve as computer vision algorithms and

hardware features advance.
OCS inspection methods continue to evolve as
technology develops. Inspection with traditional
REFERENCES
methods that are very simple requires much inspection
time and endangers the inspectors requiring more [1] L. Liudvinavičius and S. Dailydka, “The
effective methods such as contact measurement Aspects of Catenary Maintenance of Direct
methods. Contact measurement methods are often less Current (DC) and Alternating Current (AC),”
reliable and endanger railway operations, so non- Procedia Eng., vol. 134, no. Dc, pp. 268–275,
contact methods are developed using ranging 2016, doi: 10.1016/j.proeng.2016.01.007.
technology and cameras. Researchers sometimes [2] A. Oya, K. Nishi, M. Shimizu, T. Mandai, and
combine non-contact methods for specific purposes, M. Tago, “Application of overhead rigid
such as accuracy improvement and self-calibration. conductor line to mountain tunnel of
Computer vision has the potential capability to detect conventional lines,” Proc. IEEE Int. Conf. Ind.
objects and can be utilized mainly in any OCS Technol., vol. 2005, pp. 170–174, 2005, doi:
inspection activities. Sometimes additional or 10.1109/ICIT.2005.1600630.
modified methods are required depending on the [3] J. Wu, “Chapter 1 - Introduction,” in High-
inspection object type to obtain promising results. Speed Railway, J. B. T.-P. and C. L. S. Wu, Ed.
Each method has advantages and disadvantages Academic Press, 2018, pp. 1–26. doi:
depending on the parameters of the object to be https://doi.org/10.1016/B978-0-12-812886-
measured. The challenge in the future is to improve 2.00001-X.
the shortcomings of the methods that have been [4] L. Yu, S. Gao, D. Zhang, G. Kang, D. Zhan,
implemented, especially in accuracy, precision, and C. Roberts, “A Survey on Automatic
computation time cost, robust detection on more Inspections of Overhead Contact Lines by
complex backgrounds, and faster computing. This Computer Vision,” IEEE Trans. Intell. Transp.
challenge is needed to fill real-time online Syst., vol. 23, no. 8, pp. 10104–10125, Aug.
measurements to speed up the inspection process 2022, doi: 10.1109/TITS.2021.3119023.
significantly. Computer vision measurement methods [5] A. Collina, F. Fossati, M. Papi, and F. Resta,
“Impact of overhead line irregularity on

58
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

current collection and diagnostics based on the and Z. Han, “Advances of research on high-
measurement of pantograph dynamics,” Proc. speed railway catenary,” J. Mod. Transp., vol.
Inst. Mech. Eng. Part F J. Rail Rapid Transit, 26, no. 1, pp. 1–23, 2018, doi: 10.1007/s40534-
vol. 221, no. 4, pp. 547–559, 2007, doi: 017-0148-4.
10.1243/09544097F02105. [17] Dr Saeed Fararooy, Clemens Mair, “Review of
[6] S. Östlund, A. Gustafsson, L. Buhrkall, and M. Railway Overhead Wire Geometry
Skoglund, “Condition monitoring of Measurement systems,” pp. 16–19, 1998.
pantograph contact strip,” IET Semin. Dig., [18] P. M. Keen, “Monitoring overhead line
vol. 2008, no. 12216, 2008, doi: equipment,” IEE Colloq., no. 509, pp. 2–4,
10.1049/ic:20080343. 1998, doi: 10.1049/ic:19981002.
[7] M. Elia et al., “Condition monitoring of the [19] I. Aydin, M. Karaköse, and E. Akin, “A robust
railway line and overhead equipment through anomaly detection in pantograph-catenary
onboard train measurement - An Italian system based on mean-shift tracking and
experience,” IET Semin. Dig., vol. 2006, no. foreground detection,” Proc. - 2013 IEEE Int.
11575, pp. 102–107, 2006, doi: Conf. Syst. Man, Cybern. SMC 2013, pp. 4444–
10.1049/ic:20060052. 4449, 2013, doi: 10.1109/SMC.2013.757.
[8] Y. Wang et al., “Automatic Visual Inspection [20] Z. Liu, W. Liu, and Z. Han, “A high-precision
and Condition-Based Maintenance for detection approach for catenary geometry
Catenary,” Maint. Manag., pp. 0–16, 2020, parameters of electrical railway,” IEEE Trans.
doi: 10.5772/intechopen.82149. Instrum. Meas., vol. 66, no. 7, pp. 1798–1808,
[9] T. Usuda, “Prediction of Contact Wire Wear on 2017, doi: 10.1109/TIM.2017.2666358.
High-speed Railways Influential Factors on [21] Japanese National Railways, “Overhead
Adhesion between Wheel and Rail under Wet Contact System,” in Railway Electrification,
Conditions,” no. 36, p. 2011, 2011. 2nd ed., 1988, p. 81.
[10] M. Swift, G. Aurisicchio, and P. Pace, “New [22] S. Midya, D. Bormann, T. Schütte, and R.
practices for railway condition monitoring and Thottappillil, “Pantograph arcing in electrified
predictive analysis,” IET Conf. Publ., vol. railways-mechanism and influence of various
2011, no. 581 CP, 2011, doi: parameters - Part II: With AC traction power
10.1049/cp.2011.0578. supply,” IEEE Trans. Power Deliv., vol. 24,
[11] A. Takahashi, T. Kishi, and H. Yamamoto, no. 4, pp. 1940–1950, 2009, doi:
“Overhead Contact Line Monitoring and 10.1109/TPWRD.2009.2021036.
Prediction of Contact Wire Localized Wear [23] Y. Bai, J. Zhang, W. Liu, and X. Liu, “Study
Points,” JR East Tech. Rev., vol. 29, no. 29, pp. on influence of contact wire design parameters
22–25, 2014, [Online]. Available: on contact characteristics of Pantograph-
http://trid.trb.org/view/2014/C/1350994 Catenary,” IEEE ICIRT 2013 - Proc. IEEE Int.
[12] S. Liu, Q. Wang, and Y. Luo, “A review of Conf. Intell. Rail Transp., pp. 268–273, 2013,
applications of visual inspection technology doi: 10.1109/ICIRT.2013.6696306.
based on image processing in the railway [24] Menteri Perhubungan, “Persyaratan Teknis
industry,” Transp. Saf. Environ., vol. 1, no. 3, Jalur Kereta Api,” PM. No. 60 Tahun 2012, pp.
pp. 185–204, 2019, doi: 10.1093/tse/tdz007. 1–57, 2012.
[13] S. B. Gao, Z. G. Liu, and L. Yu, “Detection and [25] G. Jing, X. Qin, H. Wang, and C. Deng,
monitoring system of the pantograph-catenary “Developments, challenges, and perspectives
in high-speed railway (6C),” 2017 7th Int. of railway inspection robots,” Autom. Constr.,
Conf. Power Electron. Syst. Appl. - Smart vol. 138, no. January, p. 104242, 2022, doi:
Mobility, Power Transf. Secur. PESA 2017, 10.1016/j.autcon.2022.104242.
vol. 2018-Janua, pp. 1–7, 2018, doi: [26] M. A. Abdullah, Y. Michitsuji, and M. Nagai,
10.1109/PESA.2017.8277746. “System identification of railway trains
[14] S. Gao, “Automatic Detection and Monitoring pantograph for active pantograph simulation,”
System of Pantograph-Catenary in China’s MOVIC 2010 - 10th Int. Conf. Motion Vib.
High-Speed Railways,” IEEE Trans. Instrum. Control. Proc., no. June 2014, 2010, doi:
Meas., vol. 70, 2021, doi: 10.1299/jsmemovic.2010._2c31-1_.
10.1109/TIM.2020.3022487. [27] C. O’Donnell, R. Palacin, and J. Rosinski,
[15] D. Risti, M. Franke, and K. Michels, “A Pantograph Damage and Wear Monitoring
Review of Vision-Based On-Board Obstacle System. 2006. doi: 10.1049/ic:20060065.
Detection and,” 2021. [28] P. Boffi et al., “Optical fiber sensors to
[16] Z. Liu, Y. Song, Y. Han, H. Wang, J. Zhang, measure collector performance in the
59
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

pantograph-catenary interaction,” IEEE Sens. between pantograph and overhead con,” p. 20,
J., vol. 9, no. 6, pp. 635–640, 2009, doi: 2012.
10.1109/JSEN.2009.2020244. [40] I. Aydin, M. Karakose, and E. Akin, “Anomaly
[29] M. Bocciolone, G. Bucca, A. Collina, and L. detection using a modified kernel-based
Comolli, “Pantograph-catenary monitoring by tracking in the pantograph-catenary system,”
means of fibre Bragg grating sensors: Results Expert Syst. Appl., vol. 42, no. 2, pp. 938–948,
from tests in an underground line,” Mech. Syst. 2015, doi: 10.1016/j.eswa.2014.08.026.
Signal Process., vol. 41, no. 1–2, pp. 226–238, [41] C. J. Cho and H. Ko, “Video-based dynamic
2013, doi: 10.1016/j.ymssp.2013.06.030. stagger measurement of railway overhead
[30] S. Sung, “A study on the measurement of power lines using rotation-invariant feature
contact force of pantograph on high speed matching,” IEEE Trans. Intell. Transp. Syst.,
train,” KSME Int. J., vol. 20, no. 10, p. vol. 16, no. 3, pp. 1294–1304, 2015, doi:
1548~1556, 2006. 10.1109/TITS.2014.2361647.
[31] J. Peng et al., “Dynamic detection for the [42] P. Catenary et al., “LabVIEW application for
contact line gradient in electrified railway,” motion tracking using USB camera LabVIEW
High-Power Diode Laser Technol. Appl. VII, application for motion tracking using USB
vol. 7198, p. 71981P, 2009, doi: camera,” 2017, doi: 10.1088/1757-
10.1117/12.808602. 899X/200/1/012008.
[32] R. Puschmann and D. Wehrhahr, “Ultrasonic [43] S. R. Anghel, M. Panoiu, and C. Abrudean,
measurement of contact wire position,” Eb- “Intelligent Application for Monitoring the
Elektriche Bahnen, vol. 109, no. 7, pp. 323– Pantograph-Catenary Contact in Electric
324, 2011. Railway Transportation,” 2018 3rd Int. Conf.
[33] S. Wang, L. Niu, and N. Li, “Research on Smart Sustain. Technol. Split. 2018, no. Lc, pp.
Image Recognition of Insulators Based on 1–6, 2018.
YOLO Algorithm,” 2018 Int. Conf. Power [44] D. Zhan, D. Jing, M. Wu, D. Zhang, L. Yu, and
Syst. Technol., no. 201805150000003, pp. T. Chen, “An Accurate and Efficient Vision
3871–3874, 2018. Measurement Approach for Railway Catenary
[34] R. Ge, Y. Zhu, Y. Xiao, and Z. Chen, “The Geometry Parameters,” IEEE Trans. Instrum.
subway pantograph detection using modified Meas., vol. 67, no. 12, pp. 2841–2853, 2018,
faster R-CNN,” Commun. Comput. Inf. Sci., doi: 10.1109/TIM.2018.2830862.
vol. 685, pp. 197–204, 2017, doi: 10.1007/978- [45] Z. Huang, L. Chen, Y. Zhang, Z. Yu, H. Fang,
981-10-4211-9_20. and T. Zhang, “Robust contact-point detection
[35] L. De Donato et al., “A Survey on Audio- from pantograph-catenary infrared images by
Video Based Defect Detection Through Deep employing horizontal-vertical enhancement
Learning in Railway Maintenance,” IEEE operator,” Infrared Phys. Technol., vol. 101,
Access, vol. 10, pp. 65376–65400, 2022, doi: no. June, pp. 146–155, 2019, doi:
10.1109/ACCESS.2022.3183102. 10.1016/j.infrared.2019.06.015.
[36] M. Di Summa et al., “A Review on Deep [46] D. Zhang, S. Gao, L. Yu, G. Kang, D. Zhan,
Learning Techniques for Railway and X. Wei, “A Robust Pantograph-Catenary
Infrastructure Monitoring,” IEEE Access, vol. Interaction Condition Monitoring Method
11, no. July, pp. 114638–114661, 2023, doi: Based on Deep Convolutional Network,” IEEE
10.1109/ACCESS.2023.3309814. Trans. Instrum. Meas., vol. 69, no. 5, pp. 1920–
[37] M. Elgendy, Deep Learning for Vision 1929, 2020, doi: 10.1109/TIM.2019.2920721.
Systems. Manning Publications, 2020. [47] Y. G. Wang et al., “Construction of Visual
[Online]. Available: Inspection Database for Catenary on High-
https://books.google.co.id/books?id=6gkLzA speed Railways,” MATEC Web Conf., vol. 124,
EACAAJ pp. 1–5, 2017, doi:
[38] F. V. Sncf, “Contribution of Terrestrial Laser 10.1051/matecconf/201712401006.
Scanning for monitoring and inspection of [48] J. Chen, Z. Liu, H. Wang, A. Nunez, and Z.
railway infrastructure,” no. November, 2013. Han, “Automatic defect detection of fasteners
[39] British Standards Institution, “British on the catenary support device using deep
Standards Institution (2012) BS EN 50317: convolutional neural network,” IEEE Trans.
Railway applications. Current collection Instrum. Meas., vol. 67, no. 2, pp. 257–269,
systems. Requirements for and validation of 2018, doi: 10.1109/TIM.2017.2775345.
measurements of the dynamic interaction [49] Y. Liu et al., “A Coarse-to-Fine Detection
between pantograph and overhead contact line Method of Pantograph-Catenary Contact

60
Jurnal Perkeretaapian Indonesia (Indonesian Railway Journal) Vol.8 No. 2 Oktober, 2024
p-ISSN 2550-1127/e-ISSN 2656-8780

Points Using DCNNs,” IFAC-PapersOnLine, of railways,” Proceeding 6th Int. Conf. Pattern
vol. 52, no. 24, pp. 71–75, 2019, doi: Recognit. Inf. Process., vol. 1, no. 7, pp. 275–
10.1016/j.ifacol.2019.12.383. 287, 2001.
[50] G. Kang, S. Gao, L. Yu, and D. Zhang, “Deep [60] X. Wu, P. Yuan, Q. Peng, C. W. Ngo, and J. Y.
Architecture for High-Speed Railway Insulator He, “Detection of bird nests in overhead
Surface Defect Detection: Denoising catenary system images for high-speed rail,”
Autoencoder with Multitask Learning,” IEEE Pattern Recognit., vol. 51, pp. 242–254, 2016,
Trans. Instrum. Meas., vol. 68, no. 8, pp. 2679– doi: 10.1016/j.patcog.2015.09.010.
2690, 2019, doi: 10.1109/TIM.2018.2868490. [61] R. Tian, H. Shi, B. Guo, and L. Zhu, “Multi-
[51] P. Tan et al., “Mask R-CNN and multifeature scale object detection for high-speed railway
clustering model for catenary insulator clearance intrusion,” Appl. Intell., vol. 52, no.
recognition and defect detection,” J. Zhejiang 4, pp. 3511–3526, 2022, doi: 10.1007/s10489-
Univ. Sci. A, vol. 23, no. 9, pp. 745–756, 2022, 021-02534-9.
doi: 10.1631/jzus.A2100494. [62] S. Kovalskyi and V. Koval, “Comparison of
[52] A. Landi, L. Menconi, and L. Sani, “Hough image processing techniques for defect
transform and thermo-vision for monitoring detection,” 2024.
pantograph-catenary system,” Proc. Inst. [63] T. Lei and A. K. Nandi, Image Segmentation.
Mech. Eng. Part F J. Rail Rapid Transit, vol. Wiley, 2022. doi: 10.1002/9781119859048.
220, no. 4, pp. 435–447, 2006, doi: [64] H. Feng, Z. Jiang, F. Xie, P. Yang, J. Shi, and
10.1243/0954409JRRT41. L. Chen, “Automatic fastener classification
[53] I. Aydin, M. Karakose, and E. Akin, “A new and defect detection in vision-based railway
contactless fault diagnosis approach for inspection systems,” IEEE Trans. Instrum.
pantograph-catenary system,” Proc. 15th Int. Meas., vol. 63, no. 4, pp. 877–888, 2014, doi:
Conf. Mechatronics, MECHATRONIKA 2012, 10.1109/TIM.2013.2283741.
2012. [65] P. Somwang and E. Muangklang, “Image
[54] T. Koyama, M. Ikeda, K. Nakamura, S. Processing for Quality Control in
Tabayashi, and M. Niwakawa, “Measuring the Manufacturing Process,” in 2019 16th
contact force of a pantograph by image International Conference on Electrical
processing technology,” WIT Trans. Built Engineering/Electronics, Computer,
Environ., vol. 127, pp. 189–198, 2012, doi: Telecommunications and Information
10.2495/CR120171. Technology (ECTI-CON), Jul. 2019, pp. 782–
[55] I. Aydin, E. Karakose, M. Karakose, M. T. 785. doi: 10.1109/ECTI-
Gencoglu, and E. Akin, “A new computer CON47248.2019.8955421.
vision approach for active pantograph control,” [66] HCL technologies, “Decoding the dichotomy:
2013 IEEE Int. Symp. Innov. Intell. Syst. Appl. Traditional Image Processing vs. Deep
IEEE INISTA 2013, 2013, doi: Learning Whitepaper,” Imaging Mach. Vis.
10.1109/INISTA.2013.6577665. Eur., pp. 1–6, 2020, [Online]. Available:
[56] I. Aydin, “A new approach based on firefly https://www.imveurope.com/sites/default/files
algorithm for vision-based railway overhead /content/white-paper/pdfs/HCL_IMVE_WP-
inspection system,” MEASUREMENT, vol. 74, ImageProcessing_vs_DL.pdf
pp. 43–55, 2015, doi: [67] J. Yang, S. Li, Z. Wang, H. Dong, J. Wang, and
10.1016/j.measurement.2015.07.022. S. Tang, “Using deep learning to detect defects
[57] M. R. Arthington, P. T. Barnes, and S. R. in manufacturing: A comprehensive survey
Duncan, “Non-contact force measurement for and current challenges,” Materials (Basel).,
current collection in a 25kV overhead line vol. 13, no. 24, pp. 1–23, 2020, doi:
electrified railway,” IET Conf. Publ., vol. 10.3390/ma13245755.
2016, no. CP701, pp. 1–6, 2016, doi: [68] I. Aydin, O. Yaman, M. Karaköse, and S. B.
10.1049/cp.2016.1206. Çelebi, “Particle swarm based arc detection on
[58] G. Karaduman, M. Karakose, and E. Akin, time series in pantograph-catenary system,”
“Deep Learning Based Arc Detection in INISTA 2014 - IEEE Int. Symp. Innov. Intell.
Pantograph-Catenary Systems,” pp. 904–908, Syst. Appl. Proc., pp. 344–349, 2014, doi:
2017. 10.1109/INISTA.2014.6873642.
[59] H. Möller, B. Hulin, W. Krötz, and B. Sarnes,
“Video based obstacle detection in catenaries

61