1. Introduction

Rail transportation demand is steadily increasing over the world, particularly in metropolitan regions with rapidly growing populations. Even in Europe, where population growth is slower, estimates show an increase in the proportion of people who travel by train. This causes National railways infrastructure maintenance and renewals costs, of consisting of about 300,000 km of combined track to exceed €25 billion per annum, across Europe [1], [2]. Harnessing data and analytics could help European rail-infrastructure operators to better target their maintenance spending more productively. In Great Britain’s rail network, there are 40,000 bridges and tunnels, 9,000 level crossings and 9,941 miles (16,000 km) of railway tracks. It has been reported [3], [4] that in Great Britain the rail network experiences 4.7 million train journeys every single day, demonstrating the importance of railway infrastructure. Railways require regular maintenance to ensure a safe operating condition. It costs over £1 billion p.a. in the UK, accounting for 18% of Network Rail’s overall expenditure [5], [6]. The high-level quantitative information relating to maintenance is depicted in figure 1. It shows the trends in total maintenance, in 2020-21 prices. Maintenance expenditure has been steadily increasing since 2013-14 and is expected to reach over 1,830 million pounds in 2020-21.

Figure 1. Total maintenance expenditure, 2011-12 to 2020-21 (2020-21 prices)

Therefore, preserving or improving the safety, reliability and quality of the whole railway system is a key challenge and paramount for passengers, employees, and the entire rail network. Without reliable rail track maintenance, safety of rail network will be at risk, and delays would occur regularly. For these reasons, innovative maintenance solutions for railway systems, as well as integration of maintenance into operation, are constantly studied and developed to ensure a better management of the railways network.

Nowadays different sectors in industry benefits of automation, which has led to the development of a number of robotic solutions to maintenance and repair applications in the industry. For instance, miniature robot models have been applied in different maintenance tasks including, highways, aircraft servicing, underwater facilities, power line maintenance [7], fault detection [8], track cleaning or performing repair job such as 3D printing [9], [10]. A wheeled robot with a manipulator can undertake a variety of dangerous and remote operational tasks, such as tunnel inspection [11] or cleaning of nuclear reactors [12]. They could also be utilised in the inspection and maintenance of railway tracks [13]. Automation and computational intelligence techniques can dramatically improve the efficiency and effectiveness of maintenance. This guideline applies to the railway maintenance industry as well [14].

Many different robots have been designed to do various railway track maintenance tasks, however the most are limited to specific scenarios or uses or applications [15]. To perform the intended tasks, autonomous robots, like all other technical systems, must meet certain requirements that vary depending on the individual application or tasks.

Autonomous systems will be an element for solving the trade-off between the transport capacity challenge and the maintenance cost and time reduction. Rail infrastructure managers from all over the world are interested in developing automatic inspection systems that can detect rail flaws, sleepers’ irregularities, and missing fastening elements as high-speed railway traffic grows. These systems can improve the ability to detect defects and minimise the inspection time, allowing for more frequent railway network maintenance. The condition of the railway track is also monitored as part of the maintenance strategy [16]. Currently, industries use equipment for inspection and maintenance activities separately; these two have not yet been merged. The introduction of artificial intelligence (AI) and cognitive analytics technologies into these two procedures can make the entire process dynamic and autonomous [15].  

In this paper, first, the role of robotic and autonomous systems in track maintenance are discussed as well as the objectives and the related maintenance techniques. Following that, a review of a number of challenges related to the autonomous maintenance vehicles are mentioned. So, in this regard, various ways of positioning on the railway track are investigated. Then sensors and their functions in localisation as well as several fusion approaches aimed at enhancing positioning accuracy on the railway track are pointed out. Finally, the uncertainty sources contributing to error in railway location systems, are highlighted.

2. Railway Maintenance objectives and the related Techniques

2.1. Maintenance policy

Rail track maintenance encompasses all technical and administrative actions aimed at inspecting, repairing, and maintaining railway tracks in order to keep trains moving smoothly and securely while also extending the service life [17], [18]. The following are the primary objectives of railway track maintenance:

1. Due to high speed of trains, heavy axle loads, and repetitive loads, the track structure's strength continues to deteriorate.
2. The track structure is subjected to various degrading factors such as rain, sunlight, and wind. The deterioration of rolling stock and rails is unavoidable.
3. The track structure has to withstand so many other curvatures, speeds and loads effects, particularly at curves, points, and crossings.

That is why it is critical to maintain railway track on a regular basis. Railway maintenance tasks, on the other hand, are costly, and poor maintenance or inability to conduct preventive maintenance, will result in severe consequences. As a result, the application of robotic and autonomous systems in this area is proposed to undertake these maintenance tasks with higher accuracy and reliability. According to the conducted research in [19], the majority of robotic and autonomous systems advances in railway inspection and maintenance area are related to the rolling-stock and rail-track with 56% and 28% respectively. The cost-effectiveness of robotics automation in railway track maintenance and related tasks have been already proven [20].

Nowadays track maintenance is achieved through the utilization of highly specialized machines such as replacement, track stabilization, ballast injection (stone blowing), sleeper replacement, tamping the ballast, excavation, spiking rail, tightening bolts, and aligning the track [21]. A number of these robots are so massive which can conduct more than one task, for instance, ballast tamping combined with track lining and leveling. However, in order to get the best result in each part, it is decided to apply specific robotic system for each task. As a result, each individual robot will concentrate on a single target for doing a high-specialised & efficient job.

In terms of railway track maintenance, a fully autonomous robotic system detect and remove sleeper bolts, feed new fastener, and assemble them in high-speed train lines [22]. Rowshandel et al. [23], [24] suggested an integrated robotic system, that comprises of a mechanized trolley, a robot, a commercially available alternating current field measurement (ACFM) device, and a laser distance sensor, for identification and characterization of surface-breaking defects with high precision in rails. Auto-Scan [25] is another example of an autonomous rail inspection system. It is an autonomous trolley that detect defects on the railway track using electromagnetic acoustic transducers (EMATs). Railpod [26], a commercially available autonomous rail inspection robotic platform, can run both on and off track, but only on plain surfaces, because it has both rail and pneumatic wheel reconfigurable mechanisms.

A robust rail condition monitoring methodology proposed in [27], in which they applied a laser scanner mounted on a moving rail vehicle to detect track fractures, scorings, and excessive wear. An autonomous track geometry diagnostician's computer system was proposed in [28], which alerts poor track locations. Missing or broken track components including bolts, clips, ties, tie plates, anchors, and turnout components have also been detected using robotic and autonomous systems [29]. For instance, a new method for detecting missing or defective rail fastener problems is proposed in research [30], which uses the histogram of oriented gradients information and a mixture of linear Support Vector Machine (LSVM) classifiers.

Besides providing the infrastructures for having an autonomous maintenance vehicle, proper maintenance of railway assets is also necessary, which should be done periodically. It can be periodic updating, periodic comprehensive maintenance, regular inspection and key repairing of track [31]. Services are at significant risk of failure if they are not properly maintained, with negative consequences for user satisfaction and asset professionals' performance. As a result, two types of maintenance techniques have been established by National Railway to keep the tracks as functional as possible: preventive and corrective railway maintenance. Both demand a significant amount of resources, specialised equipment, and well-trained personnel [32].

2.1.1. Preventive maintenance

Preventive maintenance (PM) is an important part of maintenance activity. An integral aspect of PM is systematic inspection, detection, and repair of incipient failures, either before to their occurrence or before they proceed to a failure state, by competent persons involved in maintenance [33], [34]. PM's main goals are to extend the useful life of capital equipment, reduce critical equipment breakdowns, improve the planning and scheduling of needed maintenance works, reduce production losses due to equipment failures, promote the health and safety of maintenance personnel, and provide maximum system reliability and safety with the least amount of maintenance resources [35]. Totally, PM is divided into the following options, as shown in figure 2.

1. Inspection: comparing physical, electrical, mechanical, and other properties (as appropriate) to the expected standards to evaluate the serviceability of materials/items.
2. Servicing: cleaning, lubricating, charging, preserving, and so on, of items/ materials on a regular basis to avoid incipient breakdowns.
3. Calibration: determining the value of an item's attributes on a regular basis by comparing it to a recognised standard with known accuracy.
4. Testing: testing or checking out on a regular basis to verify serviceability and discover electrical/mechanical degradation.
5. Alignment: changing the stated variable aspects of an item in order to achieve optimum performance.
6. Adjustment: periodically modifying specified material variable parts in order to achieve optimal system performance.
7. Installation: replacement of limited-life parts or equipment experiencing time cycle or wear degradation, to maintain the stated system tolerance, on a regular basis.

Figure 2. Elements of preventive maintenance

Preventive maintenance for these operations takes place at specified intervals or according to certain criteria to limit the likelihood of equipment failure or degradation in the functioning. According to the structure of a rail track, each element must be considered, as mispositioning of one of them might result in poor alignment and, in the worst-case scenario, train derailment. As a result,, it can be divided into two parts: day-to-day maintenance (scheduled continuously), and seasonal maintenance [36].

• Day-to-day maintenance:

The goal of this type of maintenance, from first inspection to final control, is to get the track operational and running as soon as possible. Three main units are mostly used for maintaining the track: Stoneblowers, Tamping, and Dynamic Track Stabilisation (DTS) [37]. These machines are employed for large-scale refurbishment projects, but a vast fleet of other units operates for smaller tasks. Stoneblowers and Tamping machines act on the ballast and the rail. These three devices allow the track to be repositioned and reinforced. The operations' purpose is to fix the track's position by shifting ballast and adjusting rails and sleepers. However, maintenance also entails the removal of defects, particularly on the rail. As previously stated, the rail is prone to many failures as a result of the train's passing and the force exerted, and if not addressed promptly, the defect might spread and lead to the disintegration of the rail [38]. The most common rail problems are head check, gauge corners, cracks, and squats, which initiate on the rail surface but spread in-depth if trains continue to run [38]. It can be kept in check by routinely grinding the rail surface and lubricating the joint between the rail and the wheels.

• Seasonal Track Maintenance:

Weather is another factor that that has a significant impact on the train's operating conditions. Temperature changes, leaves, plants, and ice can either damage the track or limit the train’s abilities. The functional operation of the track, as well as the safety of passengers, are significantly influenced in this circumstance [39]. Various operations are carried out to address this problem, depending on the weather conditions. In the winter, for instance, rails are prone to freezing and ice formation; in this situation, scraping or blowing hot air and then spraying hot liquid on the rail will prevent the rail from freezing [38]. In autumn, leaves on the rail, reduce the friction between the wheels and the rail, causing friction to increase, and the train’s wheels hardly adhere to the rail, especially during acceleration phases. In this situation, by blowing high pressured air by the Rail Head Treatment Trains (RHTT), the leaves would be taken away from it. It then adds an adhesion gel to the wheels to strengthen their grip on the rail. Workers are also meant to cut down the trees, that are suspected of colliding with the train. Plants on the railway track begin to grow in the summer, causing the track's structure to become unsettled. For this case, herbicides are sprayed on the track to kill the seeds, and vegetation is pruned back to maintain the track accessible. Another issue in the summer season is temperature. As rails are thermal conductors, they retain the heat. As a consequence, it expands and is no longer in its proper position [38]. So, trains are subject to derailment. To prevent this failure, some parts of the rail are painted in white to reflect the sun's rays. However, in other circumstances, preventing expansion is insufficient. The rail tension is adjusted to allow the track to dilate and stretch without buckling.

2.1.2. Corrective maintenance

Corrective maintenance (CM) is the task of locating, isolating, and resolving a fault so that a failed equipment can be substituted or restored to an operational condition within the tolerances or limits set for in-service operations [33]. It plays a significant role in the efficiency of maintenance organisations. Sometimes, the track is too damaged to be restored or upkeep may be too costly [32]. In this case, they proceed to corrective maintenance which consists in replacing the degraded pieces [34]. Concerning railway maintenance, two types of operations are carried out depending on the damage level. If only a small portion of rail requires repair, for instance, replacement of a small part, manual operations are done, including cutting out the defective rail part, removing the rail part from the site using the road-rail vehicle, clearing the space around, bringing the new rail, proceed to illuminate thermic welding, proceed to profile and grinding, inspect the final rail, are performed [40]. But for more advanced maintenance, specifically designed machines conduct the tasks.

The CM process takes time to be achieved and requires the use of multiple machines as well as the participation of several personnel. The activities for larger maintenance are almost entirely automated. Track renewal is done when the rails have sustained too much damage. From the dismantling of the previous rails through the inspection of the new track, a special purpose machine is employed to assure the complete replacement of the track [41]. Totally, CM may be classified into five major categories including: fail-repair, salvage, rebuild, overhaul, and servicing as shown in figure 3:

Figure 3. Elements of corrective maintenance

Overall, maintenance is critical in the railway industry since it ensures both safety and profitability. In the railway industry, there has been a tendency toward increased automation and productivity. As a result, it has become larger, more expensive, and more technically complicated than ever. The track infrastructure is a significant part of the railway system, and its upkeep is a key financial consideration in technical, administrative, and managerial decisions.

Now, in order to do autonomous maintenance, it necessitates to first identify the location of the railway vehicle precisely. So, in the next section, possible approaches are investigated.

3. Railway vehicle localisation strategies

Railway vehicle positioning plays a prominent role in the safety of the system since the positioning information is used for train separation and control. Rail vehicles are confined to travel along the railway and an error in train location might result in a dangerous overestimation of braking distance. So, an accurate and reliable estimation of the location of rail vehicles such as trains, trams, subways, and rail-robots, is critical to rail-system management [42], [43]. A wide range of sensors and infrastructures have been proposed and implemented, for railway vehicle positioning, so their classification is required as well. These sensors are mostly divided into two classes: 

• Elements in the railway environment (infrastructure-based)
• On-board sensors (infrastructure-less)

3.1. Track infrastructures-based strategy

Infrastructures in the railway environment are used as a way for determining the location of trains on the railway track. Examples of track-side sensors are magnetic coils, cable loops, contacts, track-circuits, axle counters, transponders and radio Balises [44], [45]. A number of them are shown in figure 4. The presence of a train on rail tracks, for example, is detected using a track circuit (a simple electrical device). This equipment is not devoted to locating the train specifically but rather to locate it indirectly on a track portion. The location can also be determined using detectors installed along the railway, which are used for train protection. These sensors could be transponders (Balises), which communicate with the train on-board equipment, when the train passes over them [46]. At best, this technology can provide train position information with a precision of a few hundred metres, which is sufficient for providing a train safety system, especially when there is a risk of collision between them. The collision avoidance operations are accomplished by the protection systems with interlocking [42], which specify whether a train can access a track section/ block, or not. As a result, any future obstruction on the railway track is avoided. However, the main drawback of these localisation solutions is installation and the high cost of maintenance. Because of this, these infrastructures are usually spread along the railway track, with distances ranging from tens of metres for current Radio-Frequency Identification (RFID)-based systems to tens of kilometres for conventional commercial transnational railway systems [42]. In terms of application, Balises or cable loops are mostly used in continuous train control systems, such as the European Train Control Systems (ETCS) Level 2 or the German LZB (Linienzugbeeinflussung) with the goal of better localising the train [47]. Currently, localisation strategies used, for example, in subway applications, heavily rely on ground sensor infrastructure such as Balises/tags and beacons [48], [49].

Figure 4. Track-side infrastructures

3.2. On-board sensors

On-board sensors are including tachometers, inertial sensors, satellite-based positioning systems, and other on-board sensors [50]. Several on-board sensors are provided in figure 5. In terms of train positioning based on these sensors, all the components and computations parts are onboard, and in contrast to infrastructure-based localisation methods, it requires a map of the railway tracks as well [51]. The track map contains the tracks information, their connections, and measurable features of the track, which provides a reliable position estimation. There are multiple approaches for train localisation based on onboard sensors and a map, which are different in terms of sensor types or combinations, processing methods and evaluation scopes [51]. A combination of Global Positioning System (GPS), Doppler measurements and a track map is studied in [52].

Figure 5. On-board sensors: (a) Stereo depth camera; (b) Environmental camera; (c) IMU; (d) RTK-GPS; (e) 3D Lidar.

Lots of research have been done in the field of train positioning method based on GPS. For instance European Train Control systems (ETCS) level-3 use GPS especially for train integrity confirmation [53], [54]. Due to the train’s environment which is complex, using only GPS cannot afford reliable position data in some specific scenarios like tunnels, hilly regions, and urban canyons [55]. Therefore, multi-sensor fusion positioning methods have been proposed in this regard [56], [57]. Fusion of GPS and Inertial Navigation System (INS) is widely used as a localisation system [58]. This system works properly when there is a continuous correction by GPS, otherwise there is cumulative position error caused by INS. In other words, if there is no GPS correction, using the Inertial Measurement Unit (IMU) alone for a long time will result in a huge offset. Therefore, more sensors and information sources are necessary to precisely and consistently locate the train, such as odometers, eddy-current sensors, Doppler velocity sensors and accelerometers, digital maps and wayside transponders like RFID or Balises and other on-board equipment [59]. The necessity to augment GPS with other measurements for operational and safety reasons is discussed in [60], [61].

A multi-sensor scheme that collects data from various sensors installed on the vehicle (specifically, an IMU and a GPS) and performs a Kalman-based filtering recursion, is investigated in [62]. They focused on the solutions  that could be used on every rail vehicle, regardless of the ground equipment on the specific lines. In another research [63], an adaptive  multi-sensor  data  fusion  technique  for  the  precise assessment of  the  train’s  position  and  velocity, based on three on-board sensors namely longitudinal accelerometer, odometer and GPS receiver unit is suggested. A localisation technique for the railway vehicles with the aim of performance enhancement in terms of speed and position estimation accuracy, based on the fusion of tachometer and IMU is introduced in [64]. This fusion was thanks to the Kalman Filter (KF) theory. In another research [65], a particle filter-based localization approach for a rail-guided robot is presented. A particle filter was considered to integrate odometry with inertial measurements, laser scans and image data. So as a result, a rail map, a motion model and a perception model were developed to implement a 1D estimation.

Dead-reckoning systems such as eddy current sensors [66], [67], Doppler radar [68], Inertial Measurement Unit (IMU), and optical imaging [69], are applied for vehicle positioning based on estimating distance and direction of travel from a known fixed point. However, these systems lead to uncertainty due to inherent cumulative errors which are typically caused by wheel slippage, wear of wheels and mechanical parts, bias, and hysteresis, etc. [70]. So, these sensors cannot be utilised alone for long periods of time in safety-critical applications, such as collision avoidance and train automation [71] and need to be reset periodically to improve localisation accuracy. For this reason, whether they are used alone or in combination, necessitates to be merged with GPS [72] or track-side markings such as RFID-type devices or Balises [68]. The fusion of GNSS, inertial sensors, and odometry was examined in [73].

It is worth mentioning that current train-borne localisation systems using GNSS, odometers, and track maps, have severe shortcomings concerning accuracy and reliability. The problem is that they cannot always determine immediately that the train is situated on which of several parallel tracks. This is the most important prerequisite for the safety of railway vehicle control systems [74]. Therefore, lidar sensors as the most promising choice to complement those systems, proposed in train localisation [75]. It is used for identifying large structures and environmental changes (e.g., exiting a tunnel or entering a station), based on topological map of those features. It can be said that in the case of train positioning, lidar is related to the topological landmark detection, such as the work done in [44]. This information is then utilised to verify other sensors' location estimates and reset dead-reckoning errors.

Onboard train localisation from the perspective of safety assessment, was investigated in [45]. The researchers attempted to answer the question of how an onboard train localisation system has to be designed, that ensures safe operations. They propose a generic approach that isn't reliant on specific sensing devices. The system is based on three sources of information including a GNSS receiver, a velocity sensor, and a digital track map. Figure 6 represents the components of an onboard localisation system that fuses the incoming information from the onboard sensors. It determines the position of railway vehicle without relying on any trackside aids.

Figure 6. Main components of a railway vehicle positioning (adapted from [63])

In another classification, Durazo Cardenas et al. [55] cited the positioning sensors based on the principles of the sensor’s technology. They gathered four types of general location systems including: GNSS [76], Radiolocation [77], Proximity [78], [79] and dead-reckoning systems. A summary of these sensors can be found in [55].

In table 1, a number of on-board sensors and track-side infrastructure are compared with each other in terms of absolute and relative positioning, rate of frequency, long-term and short-term baseline, outage difficulties, and environmental impact. In this table, absolute localisation refers to a global localization solution which relies on the GNSS constellations or landmarks to restore the position and orientation information with regard to a global reference frame [80]. Relative localization refers to a local localization technique which uses onboard sensors and kinematic models to estimate the robot’s pose relative to its initial pose [80]. Baseline is defined by the distance between the rover and a reference point [81]. For instance, in terms of applying GPS, base line is about the vehicle and the base station. The length of the baseline varies between, short, medium and large [81]. Another parameter in this table is outage issue, which refers to unavailability of a sensor for a period of time, for different reasons. For example, GPS has outage issue inside the tunnels, near the buildings, or in urban canyons, and so on [82]. The last parameter is environment impact that it includes any changes in the environment such as rain, snow, fog, or any natural change in the environment and so on.

Table 1. Train Positioning Sensor Characteristics [adapted from [50]]

In the following section, first the application of the most used on-board sensors and track-side equipments, in railway system, is provided, and then the main functions, as well as their usual rate of frequency, and advantages and disadvantages are investigated.

4. Sensor hardware

As it is mentioned, one of the most important challenging tasks within any autonomous driving framework is localisation from data collected in real time. So, accurate and reliable measurement of on-board systems plays a critical role for the vehicles moving on the railway track. There are various sensors for measuring position and speed of the railway vehicles. The most common ways are including: tachometers, transponders, Balise, INS, Doppler effect and GPS [70]. A comparative survey highlighted benefits and drawbacks associated to different sensor types [83], [84]. These sensors can be evaluated and classified according to several parameters such as cost, accuracy, reliability, sensitivity, coverage, speed of response, and availability [70]. Accurate and reliable odometry information may be achieved by using a number of these sensors whose redundant or complementary data are combined intelligently to produce more accurate and reliable information. In the following, each sensor is described briefly.

• Tachometers:

Wheel angular speed sensors are widely diffused in railway applications due to their resilience and reliability. They are frequently employed as a principal form of speed measurement equipment. Through this type of sensor, when pure rolling conditions occur, the train speed can be simply estimated [85]. Various types of tachometers have been developed and applied including optical, capacitative, active, and passive electromagnetic tachometers; however, incremental optical tachometer is more accurate and efficient than the other types. Tachometer’s accuracy usually impacted by a variety of sources of errors including noises in mechanical and electrical parts, mechanical imperfections, wheel slip and slide, sampling frequency, and alteration in the wheel diameter as a result of wear and turning [70], [86]. For correcting the position inaccuracy caused by slip and slide, a number of approaches have been proposed and carried out including correction by marginal distance, mutual correction of numerous axles, and frequent resetting of the position using transponders [87].

• Transponders:

Passive and active transponders are track based items that are used in conjunction with on-board odometers (integrating tachometers), representing a reliable method for measuring the position of a train [70]. They have been used by numerous railway operators. The transponders, transmit a signal to the train-based receiver that includes information on their position and in some circumstances signalling information. Increasing the number of transponders along the track to reduce the positioning error, results in higher costs and lower reliability in the system [70].  

• Balise:

Balise is an electronic beacon or transponder that is installed between railway tracks as part of an automated train protection (ATP) system. It provides the train a position reference as well as direction information. It is an example of transponder fixed on the track to correct the position uncertainty that builds up within the train location subsystem over the time. Balises, as magnetically coupled transponders, do not require a steady energy source [88], and can be termed as passive. The Balises are placed at approximately regular intervals, that the distance between them are determined based on two factors, the speed of the trains and complexity of the railway [89]. As it is important for the trains to stop precisely at stations, point zones and buffer stops,  up to 8 Balises is required at a minimum interval defined by the design [89]. In order to increase the possibility of Balise detection by train and decrease the possibility of train missing to read a Balise, using of more than one Balise, in critical stopping locations, is considered.

• Doppler Radar:

Based on the principle of Doppler frequency shift effect, Doppler radar can calculate the train's immediate speed by analysing the frequency difference between the radar transmitted and reflected wave [90]. It is a non-contact sensor with two microwave antennas that gives accurate and consistent results independent of reflecting surface or vibration [91]. Compared to a tachometer, Doppler radar is found to provide more accurate data [92]. Heide et al. [93] by doing a number of experiments, demonstrated that the use of 24GHz coded Doppler radar can provide high precision data (within 20db) for vehicle position and speed measurement. On the other hand, Malvezzi et al. [94] while discussing odometric estimation for train protection system, stated that Doppler radar output is often affected by noise and systematic errors. Mirabadi et al. [70] also stated this issue, and identified the sources of errors due to very smooth reflective surface, and change in radiation angle owing to acceleration and braking action, vibration and bias error.

• Inertial Navigation Systems (INS):

INSs are navigational systems capable of measuring the acceleration, speed and position of a moving train along the stable axes [80]. An INS is a system that basically composed of at least three gyros and three accelerometers to derive a navigation solution. Accelerometer will measure the acceleration of the vehicle by integrating the acceleration signals both speed and position data. On the other side, Gyroscopes are used in order to measure an angular rotation of the vehicle [95]. It can be used to obtain accurate information of the trajectory of the train in a horizontal and vertical direction. Unlike tachometers which depend on wheel rotations, the INS system is self-contained, and this is the major advantages of this system. They do not need a line of sight, such as GPS. They can be used in any weather condition and environment, both underground or over ground.      

• Global Positioning System (GPS):

GPS is a satellite-based radio navigation technology which is the core functionality for any navigation system [80] and provides the absolute position information with a known ratio of error. The fundamental advantage of GPS is its long-term stability and its resistance to the accumulation of errors over time. GPS is mainly utilized for more than simple outdoor navigational tasks and it is effective in areas with a clear view of the sky. But GPS sensors are ineffective in certain areas such as tunnels, forests, underground, and underwater spaces [96]. They also have outages caused by satellite signal blockage, occasional high noise content, multipath effects, low bandwidth, and interference or jamming. Common GPS sensors are utilised for positioning, which has an accuracy of 10 m. This does not provide sufficient accuracy [97] for train localisation.  

• Light Detection and Ranging (LiDAR):

The lidar sensor is an optical device which uses laser light pulses to gather information from surfaces in the form of “points” (3D coordinates). Compared with cameras, the lidar sensor operates more reliably at different weather conditions and is less influenced by the lighting or weather conditions due to the infrared laser that provides an adequate illumination. So, even in tunnels or under bridges appropriate measurements can be obtained [74]. LiDAR sensor is also widely deployed in railway applications for different reasons, including object detection and collision avoidance, in level crossings for detecting whether there are passenger cars, trucks, or people [98]. In [75] the application of LiDAR in train-borne localisation system, investigated. It is mentioned that position measurements at turnouts remain ambiguous. Because the combination of GPS, velocity sensors, and digital track map are not capable to address the challenges of unavailability of GPS in some parts of the environments, and the failure of velocity sensors in specifying which branch at a turnout is taken by the train. In [99], a 2D Lidar is deployed for underground railway environment for specifying high-speed train localisation.

• Visual sensor:

There are various publications related to the deployment of vision sensors in railway inspection and maintenance applications such as detection of missing bolts, railhead wear, and other surface geometry inconsistencies [100]. A vision system consisting of a monocular thermal camera mounted on a train for detecting the rails in the imagery as well as for detecting anomalies on the railway, is pointed out in [101]. In another research, a prototype system for railway object detection, installed in the cab of a train, is presented in [102]; this system consists of a single camera that acquires images of the railway scene in front of the train and a near infrared laser primarily used to add illumination to the area the camera is directed at when the light is insufficient. A summary of the publications describing traditional computer vision approaches are described in [103]. The advantages of cameras, over other active sensors, are high data density and visual information that enables the detection of the boundaries of objects and the classification of these objects precisely.

In table 2, some of the positioning sensors including both Onboard sensor (infrastructure-less) and Elements in the railway environment (infra-structure-based), with their functions, advantages and disadvantages are mentioned.

Table 2. Advantages and disadvantages of sensor used in rail applications

5. Sensor fusion

5.1 Sensor fusion techniques

Sensor fusion is an essential aspect of most autonomous systems, therefore various types of algorithms and methodologies have been widely researched in recent years and are now well-established in the literature. Due to the diverse and varieties of proposed fusion algorithms in the literature, getting the current state-of-the-art fusion techniques and algorithms is a demanding task, according to a recent study [110]. Recently, several reviews on the topic of multi-sensor fusion have been published, with some describing the architectural structure and sensor technologies in AV [111], [112], others focusing on processing stages such as sensor calibration, state estimation, object and tracking [113], [114], or detailing techniques for multi-sensor fusion, such as deep learning-based approaches [115], [116].

A review study in [115] divided these techniques and algorithms into two categories: classical sensor fusion algorithms and deep learning sensor fusion algorithms. On the one hand, classical sensor fusion algorithms, such as knowledge-based approaches, statistical methods, probabilistic methods, and so on, fuse sensor data using theories of uncertainty from data flaws, such as inaccuracy and uncertainty [117]. Deep learning sensor fusion techniques, on the other hand, entail the creation of numerous multi-layer networks that enable them to process raw data and extract features in order to accomplish difficult and sophisticated tasks, such as object detection. Deep learning is a subset of artificial intelligence and machine learning that can be considered as an advancement of neural networks [115]. The quantity of research into deep learning sensor fusion algorithms in autonomous vehicles (AV) has increased noticeably. Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) algorithms are among the most used in autonomous vehicles perception systems. To increase the real-time performance of object detection, reference [118] suggested an advanced weighted-mean You Only Look Once (YOLO) CNN algorithms to merge RGB camera and LiDAR point cloud data. Some other examples of deep learning-based sensor fusion algorithms are including: ResNet or Residual Networks, is a residual learning framework that facilitates deep networks training [119], SSD or Single-Shot Multibox Detector that discretizes bounding boxes into a set of boxes with different sizes and aspect ratios per feature map location to detect objects with variant sizes [120], CenterNet [121] represents the state-of-the-art monocular camera 3D object detection algorithm.

Despite of a vast research which has been done in the field of environment perception with the deep learning approaches, the application of deep learning to localization has not received the same level of attention or maturity. As a result, there is a lot of promise for using deep Learning algorithms especially RNN to improve sequential localization data. Learning algorithms may in the future provide an end-to-end deep learning localization and mapping system that avoids feature modelling and data association, reducing errors and uncertainties associated with unmodeled dynamics and imprecise modelling [115].

5.2. Sensor fusion algorithms for vehicle-based localisation on the railway track

As it is mentioned, new approaches in instrumentation, technology, engineering and, in particular, sensor fusion, have opened up new paths for achieving better reliability and accuracy in measurement. Integration of several sensors for speed and position measurement is the concept which has attracted much interest in industry and research departments [94], [122]. Fusing measurements from different independent sensors which have different kinds of input/output attributes and characteristics, will extract the best information in terms of accuracy and reliability. So, when multi-sensor data fusion techniques are included in the system, the navigation system becomes more robust. In other words, it can increase the robustness of the system against possible faults of each element, it is composed of. This can help the railway engineers in achieving higher safety.

The use of multi-sensor data fusion techniques also gained much popularity and found to be advantageous when used in intelligent transportation system [123], [124]. Integrating different position and speed sensors will give more information about the system conditions for monitoring and control tasks [70]. Combination of sensors should be chosen optimisely to cover more aspects such as availability, reliability, speed of response, cost, and accuracy of the system, besides providing a better on-board measurement system. For instance, integrating tachometer and transponder sensors, a type of primitive sensor fusion, will increase the accuracy level by re-initiating the position at some fixed points but still does not overcome errors caused by slip and slide between two transponders [125]. A localisation algorithm for increasing the accuracy of the odometric estimation, especially in critical adhesion conditions, based on sensor fusion between tachometers and inertial measurement unit suggested in [85]. A data fusion technique proposed in [63] for train localisation system, which consists of three on-board sensors namely longitudinal accelerometer, odometer and GPS receiver unit. Wang et al. [126] proposed a train positioning method which fuses vison and millimeter-wave radar data. The proposed framework includes the loop closure detection part which eliminate the cumulative error when the train detects a key position, and the radar-based odometry part which can realize the positioning of the train on the whole railway line.

In [67], a hybrid framework for locating trains travelling on track routes based on GNSS and eddy current sensor device, implemented by an Extended Kalman Filter method, is proposed. This positioning system performs a robust localisation even in the case of noise or when a sensor fault occurs. In another research [72] two different fusion approaches which use two different system models to approximate the kinematics of train-borne location system are investigated. It consists of an eddy current sensor device providing the train velocity and a GNSS delivering the absolute position. The first fusion is based on a polar coordinate system model, whereas the second fusion approach uses two cartesian system models with the ability to switch between these models. Evaluating the fusion approaches, shows that in case of GPS failures, the first fusion approach with its system model in polar coordinates can propagate the train kinematics more adequate and thus achieve a better performance than the second fusion approach.

Another common fusion approach is based on the integration of the GPS and INSs, which typically results in a drift on the estimation of the vehicle speed and position [64]. To address this problem, a filter is used aimed at decreasing the difference between the data output signals coming from INS and another one e.g., GPS [97]. Due to GPS signal degradation in some parts of the environment including indoor environments, dense forests, and near the tall buildings, it is necessary to consider alternative solutions to compensate this outage, for instance a fusion of tachometer and IMU (Inertial Measurement Unit).

Figure 7 shows an example of fusion of different sensors for the measurement of train speed and position:

Figure 7. Integration of different sensors for measuring train position and speed

After investigating sensors, their functions and various fusion approaches applied in railway vehicle positioning, the remained gap is about the uncertainty sources which impact the positioning system performance, which necessitate to take into consideration and analysed it from different aspects. In the following they are mentioned in detail.  

6. Uncertainty in railway localisation performance

Localising maintenance vehicles, which do the task of inspection and repair, on the railway track in global level, leads to specifying the defect positioning in global level as well. Most of the off-the-shelf localisation systems in railway applications, like tamping adjustment or track geometry refinement generally need localisation requirement at meter-level or several meters, while finding the absolute location (cm level) of railway vehicles, would result in specifying the defect location in absolute level. It would also help the rail-system management, as it can assure safe running of vehicles on the railway track. In this regard, accuracy and the degree of confidence of the provided position are the main concerns. So, identifying challenges, potential factors, and effects of uncertainties, which impact the location accuracy, has high importance. In the following five key elements including sensor hardware, environment, information sources, and the positioning estimators, which contributes to vehicle’s position error, are explained [127], [128].

• Method performance uncertainty:

Methods used in the position estimation, lead to positioning error in several ways. The first item is about validity or integrity of the received information. It is necessary to assure about the presence of outliers or errors before feeding these data as input to a positioning algorithm. Because the calculated positions, reflect these wrong data and deteriorate the accuracy of the system [31]. Hence, input validation should happen as early as possible in the data flow, preferably as soon as the data is received from sensors.

Moreover, the time-domain consistency, frequency sychronisation between multiple sources, and the timeliness of processing data are essential for the stability of localising.

Another point is about the strategy considered for data fusion [70]. There are various information fusion techniques with their advantages and drawbacks [85]. Therefore, it is important to study about the selected fusion method.

The final point is about proper setting of initial conditions and performance parameters for some of the positioning algorithms, especially those employed for data fusion. Setting up proper conditions leads to positioning algorithms work properly. Otherwise, an incorrect selection results in increasing the resulting error instead of reducing it. For instance, in the extended Kalman filter algorithm, proper specification of process-noise-covariance and initial-estimate-covariance can provide an optimal gain which based on that it assigns a weight to the current state measurement and the prediction state. Bad specification will show a positioning error in the result.

• Sensor hardware uncertainty:

To guarantee a reliable basement, the hardware systems must be checked physically before using them. Since any issues including receiver failures, calibration failure, or any failure in any of the pieces of the location system, e.g., breakdown of receivers, are considered as sources of error, which can cause severe problems when used in a real system [129]. Sensor calibration is one of the main items which should be checked first, as it minimises any measurement uncertainty and verifies the precision and reproducibility of measurement instruments. Another point is about appropriate selection of sensors based on their characteristics and the requirements of the application. For instance, vision sensors such as ASUS stereo cameras are not suitable for outside environment, as they cannot work properly with strong sunshine. The output appears to be noisy from the robot’s perspective as if subject to random error, and the values obtained from the ASUS camera will be unusable [130]. Illumination dependency is only one example of the apparent noise in a vision-based sensor system which need to be taken into consideration before sensor selection. Picture jitter, signal gain, and blurring are all additional sources of noise, that potentially reducing the useful content of a color image. In terms of railway applications, sensor specification such as resolution, frequency, and sensing range should be considered as well. For instance, max sensing range for a stereo camera is 10m, it might fail at detection within fast speed of the vehicle; Fast speed will cause blur motion problem in camera imaging.

Another factor is sensor aliasing that, even with applying noise-free sensors, the amount of information is generally insufficient to identify the vehicle’s position from a single percept reading. So, techniques must be employed that base the vehicle’s localization on a series of readings and, thus, sufficient information to recover the robot’s position over time [130].  

• Pre-process uncertainty:

The number and quality of the information sources used for position estimation, have a direct impact on the positioning estimation error as well. It is necessary to provide real-time, reliable, and precise information for train positioning as it has a direct impact on vehicle safety and control, collision avoidance, and autonomous driving systems. One of these sources is the transmission rate of each technology. Because based on the strategy followed by the position estimator, ranges of some received signals can be obsolete by the time they are used. Furthermore, time synchronisation of the received information is essential, as it can cause an offset in the detection of the signal and add extra error, especially in time related positioning algorithms [15]. Sensor fusion would be a challenge in this step. For instance, to address this challenges, a novel fusion approach proposed in [131]. This method is based on the Kalman filter, which can handle asynchronous data.

Moreover, the intrinsic accuracy related to each technology, makes the position error to range from a few meters in the case of using GNSS, to several hundred meters, e.g., Global System for Mobile Communications-Railway (GSM-R). So, based on the nature of the sensors, and their features and functions, use of stand-alone or fusion of them can be considered.

Position data, generally from GPS, has varying levels of accuracy, and data from different sensors are often not well aligned spatially. This misalignment happens because of GPS position error, such as multipath reception, which is especially problematic when the GPS receiver does not have a clear view of the sky. In this regard, an alignment strategy as a pre-processing step, proposed to mitigate GPS error [131].

• Environment uncertainty:

Environment is another factor that impact the error related to the location of a system especially on coverage and accuracy. In the case of railway vehicle localisation, the environment would be a changing factor as the vehicle move from cities to open areas, where existing infrastructure will be different. So, the environment would result in different error ratio in vehicle’ location, while information sources and estimation algorithms, are the same. The estimated position can be more or less accurate depending on the number and the geo-localisation of the infrastructure in the environment. These infrastructures refer to the number of satellites in line-of-sight with the receiver as well as track-side objects, such as the Balises [89].

Based on the features of the terrain or the orography, the coverage and accuracy of the different technologies will be limited. For instance, ingress to tunnels, dense forests, tall buildings, and passage through deep and narrow track openings significantly degrade or even block the reception of the signals temporarily. The best example in this regard is about the GPS signals which may not always be available in some parts of the environment such as tunnels, hilly regions, canyons, and so on [132]. It might happen that the signals are also available, but environmental features including satellite signal blockage, occasional high noise content, multipath effects, low bandwidth, and interference or jamming, would impact the accuracy of the position estimate.

The main causes of inaccuracy and errors that affect the estimation of a position are shown in figure 8:

Figure 8. Fishbone diagram showing factors contributing to Location errors

7. Conclusion

Periodically inspection and maintenance of railway assets in railway track assets are critical to infrastructure safe operation and management to deal with continual degradation and ageing of the assets. Autonomous, and reliable methodologies can facilitate cost effective and efficient asset management. Because of the rapid progress made by robotics and autonomous systems in railway maintenance sector, developing intelligent asset management strategies for digitalisation and smart management for rail infrastructure, is a path towards the intelligent industrial 4.0. The precise, and real-time rail vehicle localisation is essential to the robot command & control, task execution, safety and efficiency. As a result, the current study gives a comprehensive overview of the hardware and software methodologies required for autonomous vehicle positioning on the railway track for maintenance purposes. In the following the main topics which covered in this review paper are mentioned:

• First, the railway infrastructure maintenance requirements and strategies, the maintenance objectives and the general preventive and corrective workflows are reviewed, which reveals that the accuracy in localisation is essential for autonomous inspection and repair systems.
• Secondly, a review of the most recent and relevant railway vehicle positioning approaches, based on infrastructures in railway environment and onboard sensors, with their principles, advantages and disadvantages are highlighted. It is identified that applying trackside positioning strategy, not only lack efficiency and accuracy for real-time applications, but also require large civil investment for construction and successive maintenance.
• Then, for obtaining a comprehensive perception for accurate localisation, the sensor fusion techniques and algorithms are discussed to review the applicability of different sensing methods. The most recent fusion approaches based on machine learning are also discussed. It is also mentioned deep learning fusion approaches are mostly applied in perception part, and further research in pose and depth estimations, loop closure detection, and feature descriptors, are needed to achieve the maturity in localisation and mapping part.
• Furthermore, the uncertainty sources in railway vehicle positioning are discussed to address the challenge features from different source, which impact the localisation accuracy and reliability. Each uncertainty sources are separately investigated and the solutions and strategies are also provided to mitigate the impact of that source.

This review provides a comprehensive discussion of the challenges of localisation technologies for railway maintenance vehicle. It provides an overall reference for localisation system architecture design for both autonomous system and manual railway rolling stocks.

Author Contributions: “Conceptualization, M.R. and H.L.; methodology, M.R. H.L.; validation, M.R., I.S.D., and A.S.; formal analysis, M.R. and H.L.; investigation, M.R.; resources, M.R.; writing—original draft preparation, M.R.; writing—review and editing, H.L., I.S.D., and A.S.; supervision, A.S.; project administration, A.S. A.H. R.A.; All authors have read and agreed to the published version of the manuscript.

Funding: This project has received funding from the Shift2Rail Joint Undertaking (JU) under grant agreement No 881574. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and the Shift2Rail JU members other than the Union.

Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Not applicable.

Data Availability Statement: No data are associated with this article.

Acknowledgments: This work was supported in part by the Centre for Life-cycle Engineering and Management at Cranfield University, in part by Network Rail Limited.

Conflicts of Interest: The authors declare no conflict of interest related to this work.

References

[1]         “Using analytics to get European rail maintenance on track | McKinsey.” https://www.mckinsey.com/industries/public-and-social-sector/our-insights/using-analytics-to-get-european-rail-maintenance-on-track (accessed Mar. 17, 2022).

[2]         “REPORT FROM THE EIM-EFRTC-CER WORKING GROUP ON MARKET STRATEGIES FOR Track Maintenance & Renewal FOLLOW UP TO THE CONCLUSIONS OF EC INNOTRACK PROJECT/WP5.”

[3]         T. Lidén, “Railway infrastructure maintenance - A survey of planning problems and conducted research,” Transp. Res. Procedia, vol. 10, no. July, pp. 574–583, 2015, doi: 10.1016/j.trpro.2015.09.011.

[4]         “Rail Track Maintenance: A Career Explained - Rail Futures.” https://railfutures.co.uk/rail-track-maintenance-a-career-explained/ (accessed Feb. 06, 2022).

[5]         “Cost benchmarking of Network Rail’s maintenance and renewals expenditure Annual report: year 2 of control period 6,” 2021.

[6]         “Predictive maintenance of rail infrastructure.” https://www.brunel.ac.uk/research/projects/predictive-maintenance-of-rail-infrastructure (accessed Jan. 28, 2022).

[7]         L. E. Parker, “ROBOTICS APPLICATIONS IN MAINTENANCE AND REPAIR,” 1998.

[8]         N. Mahfuz, O. A. Dhali, S. Ahmed, and M. Nigar, “Autonomous railway crack detector robot for Bangladesh: SCANOBOT,” 5th IEEE Reg. 10 Humanit. Technol. Conf. 2017, R10-HTC 2017, vol. 2018-January, pp. 524–527, Feb. 2018, doi: 10.1109/R10-HTC.2017.8289014.

[9]         J. James, J. Wilson, J. Jetto, A. Thomas, and V. K. Dhahabiya, “Intelligent track cleaning robot,” 2016 IEEE Int. Conf. Mechatronics Autom. IEEE ICMA 2016, pp. 332–337, Sep. 2016, doi: 10.1109/ICMA.2016.7558584.

[10]       J. Dhiyaneswaran, M. Vimal Raja, B. Ashok Kumar, and C. B. Muthu Kumaran, “Design and fabrication of railway track cleaning bot,” Mater. Today Proc., vol. 37, no. Part 2, pp. 2677–2680, Jan. 2021, doi: 10.1016/J.MATPR.2020.08.524.

[11]       E. Menendez, J. G. Victores, R. Montero, S. Martínez, and C. Balaguer, “Tunnel structural inspection and assessment using an autonomous robotic system,” Autom. Constr., vol. 87, pp. 117–126, Mar. 2018, doi: 10.1016/J.AUTCON.2017.12.001.

[12]       A. Kitamura, T. Namekawa, K. Hiramatsu, and Y. Sankai, “Operating Manipulator Arm by Robot Suit HAL for Remote In-Cell Equipment Maintenance,” http://dx.doi.org/10.13182/NT13-A24988, vol. 184, no. 3, pp. 310–319, 2017, doi: 10.13182/NT13-A24988.

[13]       S. D. Errico, “Simulation of a railway asset maintenance robot,” vol. 44, no. 0, p. 750111, 2018.

[14]       C. D. Martland, “Analysis of the potential impacts of automation and robotics on locomotive rebuilding; Analysis of the potential impacts of automation and robotics on locomotive rebuilding,” 1987. doi: 10.1109/TEM.1987.6499032.

[15]       “Robotics and industrial AI for track maintenance - Global Railway Review.” https://www.globalrailwayreview.com/article/114137/robotics-and-industrial-ai-track-maintenance/ (accessed Mar. 16, 2022).

[16]       C. Esveld, “Modern Railway Track Second Edition.”

[17]       M. Sedghi, O. Kauppila, B. Bergquist, E. Vanhatalo, and M. Kulahci, “A taxonomy of railway track maintenance planning and scheduling: A review and research trends,” Reliab. Eng. Syst. Saf., vol. 215, p. 107827, Nov. 2021, doi: 10.1016/J.RESS.2021.107827.

[18]       “The Importance of Rail and Track Maintenance - RWB Group Blog.” https://www.rwbgroup.co.uk/the-importance-of-rail-and-track-maintenance/ (accessed Jan. 27, 2022).

[19]       R. K. W. Vithanage, C. S. Harrison, and A. K. M. Desilva, “Importance and applications of robotic and autonomous systems (RAS) in railway maintenance sector: A review,” Computers, vol. 8, no. 3, 2019, doi: 10.3390/computers8030056.

[20]       R. K. W. Vithanage, C. S. Harrison, and A. K. M. M. Desilva, “Enhance 3D Point Cloud Accuracy Through Supervised Machine Learning for Automated Rolling Stock Maintenance: A Railway Sector Case Study,” Proc. - 2018 Int. Conf. Comput. Electron. Commun. Eng. iCCECE 2018, pp. 242–246, Mar. 2019, doi: 10.1109/ICCECOME.2018.8658788.

[21]       “railroad - Track maintenance | Britannica.” https://www.britannica.com/technology/railroad/Track-maintenance (accessed Dec. 24, 2021).

[22]       M. M. Trivedi, K. C. Ng, N. Lassiter, and R. Capella, “New generation of multirobot systems,” Proc. IEEE Int. Conf. Syst. Man Cybern., vol. 4, pp. 3342–3346, 1998, doi: 10.1109/icsmc.1998.726519.

[23]       H. Rowshandel, G. L. Nicholson, C. L. Davis, and C. Roberts, “An integrated robotic system for automatic detection and characterisation of rolling contact fatigue cracks in rails using an alternating current field measurement sensor,” Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit, vol. 227, no. 4, pp. 310–321, 2013, doi: 10.1177/0954409713486778.

[24]       H. Rowshandel, “The development of an autonomous robotic inspection system to detect and characterise rolling contact fatigue cracks in railway track,” no. September, p. 175, 2014, [Online]. Available: http://etheses.bham.ac.uk/4821/.

[25]       “Autonomous inspection trolley for better train track maintenance | AutoScan Project | Results in brief | H2020 | CORDIS | European Commission.” https://cordis.europa.eu/article/id/250858-autonomous-inspection-trolley-for-better-train-track-maintenance (accessed Jan. 03, 2022).

[26]       “Our Platform – RailPod, Inc.” https://rail-pod.com/?page_id=2527 (accessed Jan. 03, 2022).

[27]       “(3) (PDF) Condition Monitoring Approach Using 3DModelling of Railway Tracks With Laser Cameras.” https://www.researchgate.net/publication/314230488_Condition_Monitoring_Approach_Using_3DModelling_of_Railway_Tracks_With_Laser_Cameras (accessed Jan. 03, 2022).

[28]       J. Madejski, “Autonomous track geometry diagnostics system,” J. Mater. Process. Technol., vol. 157–158, no. SPEC. ISS., pp. 194–202, Dec. 2004, doi: 10.1016/J.JMATPROTEC.2004.09.029.

[29]       L. Wang, B. Zhang, J. Wu, H. Xu, X. Chen, and W. Na, “Computer vision system for detecting the loss of rail fastening nuts based on kernel two-dimensional principal component – two-dimensional principal component analysis and a support vector machine:,” http://dx.doi.org/10.1177/0954409715616426, vol. 230, no. 8, pp. 1842–1850, Dec. 2015, doi: 10.1177/0954409715616426.

[30]       X. Gibert, V. M. Patel, and R. Chellappa, “Robust fastener detection for autonomous visual railway track inspection,” Proc. - 2015 IEEE Winter Conf. Appl. Comput. Vision, WACV 2015, pp. 694–701, Feb. 2015, doi: 10.1109/WACV.2015.98.

[31]       Y. K. Al-Douri, P. Tretten, and R. Karim, “Improvement of railway performance: a study of Swedish railway infrastructure,” J. Mod. Transp., vol. 24, no. 1, pp. 22–37, Mar. 2016, doi: 10.1007/S40534-015-0092-0/FIGURES/11.

[32]       “Track - Network Rail.” https://www.networkrail.co.uk/running-the-railway/looking-after-the-railway/track/ (accessed Mar. 23, 2022).

[33]       B. S. Dhillon, Engineering maintenance: A modern approach. 2002.

[34]       H. Shang, C. Bérenguer, and J. Andrews, “Delayed maintenance modelling considering speed restriction for a railway section:,” http://dx.doi.org/10.1177/1748006X17709200, vol. 231, no. 4, pp. 411–428, Jun. 2017, doi: 10.1177/1748006X17709200.

[35]       S. Sharma, Y. Cui, Q. He, R. Mohammadi, and Z. Li, “Data-driven optimization of railway maintenance for track geometry,” Transp. Res. Part C Emerg. Technol., vol. 90, pp. 34–58, May 2018, doi: 10.1016/J.TRC.2018.02.019.

[36]       F. Comino, “Cobots for maintenance in the railway industry,” p. 2018, 2018.

[37]       “Track treatment fleet - Network Rail.” https://www.networkrail.co.uk/running-the-railway/looking-after-the-railway/our-fleet-machines-and-vehicles/track-treatment-fleet/ (accessed Dec. 24, 2021).

[38]       D. F. Cannon, K. O. Edel, S. L. Grassie, and K. Sawley, “Rail defects: an overview,” Fatigue Fract. Eng. Mater. Struct., vol. 26, no. 10, pp. 865–886, Oct. 2003, doi: 10.1046/J.1460-2695.2003.00693.X.

[39]       “Seasonal track treatment and weather support fleet - Network Rail.” https://www.networkrail.co.uk/running-the-railway/looking-after-the-railway/our-fleet-machines-and-vehicles/seasonal-track-treatment-and-weather-support-fleet/ (accessed Jan. 04, 2022).

[40]       “Track - Network Rail.” https://www.networkrail.co.uk/running-the-railway/looking-after-the-railway/track/ (accessed Dec. 24, 2021).

[41]       “High Output machines - Network Rail.” https://www.networkrail.co.uk/running-the-railway/looking-after-the-railway/our-fleet-machines-and-vehicles/high-output/high-output-machines/ (accessed Dec. 24, 2021).

[42]       O. Heirich, “Localization of Trains and Mapping of Railway Tracks,” 2020.

[43]       “(PDF) Requirements for Safety Relevant Positioning Applications in Rail Traffic – A Demonstrator for a Train Borne Navigation Platform Called ‘DemoOrt.’” https://www.researchgate.net/publication/224797887_Requirements_for_Safety_Relevant_Positioning_Applications_in_Rail_Traffic_-_A_Demonstrator_for_a_Train_Borne_Navigation_Platform_Called_DemoOrt (accessed Feb. 11, 2022).

[44]       T. Albrecht, K. Lüddecke, and J. Zimmermann, “A precise and reliable train positioning system and its use for automation of train operation,” IEEE ICIRT 2013 - Proc. IEEE Int. Conf. Intell. Rail Transp., pp. 134–139, 2013, doi: 10.1109/ICIRT.2013.6696282.

[45]       M. Lauer and D. Stein, “A Train Localization Algorithm for Train Protection Systems of the Future,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 2, pp. 970–979, Apr. 2015, doi: 10.1109/TITS.2014.2345498.

[46]       M. Hutchinson et al., “Precise and reliable localization as a core of railway automation ( Rail 4 . 0 ) To cite this version : HAL Id : hal-01878793 Precise and reliable localization as a core of railway automation Marais , Juliette Masson , Émilie Abstract :,” pp. 0–10, 2018.

[47]       European Railway Agency, “Railway Safety Performance in the European Union 2008,” Office, p. 44, 2008.

[48]       W. G. Temple, B. N. Anh Tran, B. Chen, Z. Kalbarczyk, and W. H. Sanders, “On Train Automatic Stop Control Using Balises: Attacks and a Software-Only Countermeasure,” 2017, doi: 10.1109/PRDC.2017.52.

[49]       J. Zhou, H. Xiao, W. Jiang, W. Bai, and G. Liu, “Automatic subway tunnel displacement monitoring using robotic total station,” Meas. J. Int. Meas. Confed., vol. 151, p. 107251, 2020, doi: 10.1016/j.measurement.2019.107251.

[50]       J. Otegui, A. Bahillo, I. Lopetegi, and L. E. Diez, “A Survey of Train Positioning Solutions,” IEEE Sens. J., vol. 17, no. 20, pp. 6788–6797, 2017, doi: 10.1109/JSEN.2017.2747137.

[51]       O. Heirich, B. Siebler, S. Sand, A. Lehner, and O. G. Crespillo, “Measurement methods for train localization with onboard sensors,” 2020 Eur. Navig. Conf. ENC 2020, 2020, doi: 10.23919/ENC48637.2020.9317435.

[52]       C. Fouque and P. Bonnifait, “Matching raw GPS measurements on a navigable map without computing a global position,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 887–898, 2012, doi: 10.1109/TITS.2012.2186295.

[53]       J. Marais, J. Beugin, and M. Berbineau, “A Survey of GNSS-Based Research and Developments for the European Railway Signaling,” IEEE Trans. Intell. Transp. Syst., vol. 18, no. 10, pp. 2602–2618, Oct. 2017, doi: 10.1109/TITS.2017.2658179.

[54]       H. Hynek and M. Mocek, “by by Galileo as an instrument of unification Galileo as an instrument of unification of the European railway transport of the European railway transport Content.”

[55]       I. Durazo-Cardenas, A. Starr, A. Tsourdos, M. Bevilacqua, and J. Morineau, “Precise vehicle location as a fundamental parameter for intelligent selfaware rail-track maintenance systems,” Procedia CIRP, vol. 22, no. 1, pp. 219–224, 2014, doi: 10.1016/j.procir.2014.07.002.

[56]       J. Wang, D. Liu, W. Jiang, and D. Lu, “Evaluation on loosely and tightly coupled GNSS/INS vehicle navigation system,” Proc. 2017 19th Int. Conf. Electromagn. Adv. Appl. ICEAA 2017, pp. 892–895, Oct. 2017, doi: 10.1109/ICEAA.2017.8065396.

[57]       K. Kim, S. Seol, and S.-H. Kong, “High-speed Train Navigation System based on Multi-sensor Data Fusion and Map Matching Algorithm,” Int. J. Control. Autom. Syst., vol. 13, no. 3, pp. 503–512, 2015, doi: 10.1007/s12555-014-0251-9.

[58]       T. Zhang and X. Xu, “A new method of seamless land navigation for GPS/INS integrated system,” Measurement, vol. 45, no. 4, pp. 691–701, May 2012, doi: 10.1016/J.MEASUREMENT.2011.12.021.

[59]       B. Stadimann, “Automation of operational train control on regional branch lines by a basic train control,” IEEE Conf. Intell. Transp. Syst. Proceedings, ITSC, pp. 50–54, 2006, doi: 10.1109/itsc.2006.1706717.

[60]       F. G. Toro, U. Becker, D. E. D. Fuentes, D. Lu, and W. Tao, “Accuracy Analysis for GNSS-based Urban Land Vehicle Localisation System,” IFAC-PapersOnLine, vol. 49, no. 3, pp. 191–196, Jan. 2016, doi: 10.1016/J.IFACOL.2016.07.032.

[61]       R. Mázl and L. Přeučil, “Sensor data fusion for inertial navigation of trains in GPS-dark areas,” IEEE Intell. Veh. Symp. Proc., pp. 345–350, 2003, doi: 10.1109/IVS.2003.1212934.

[62]       D. Selvi, E. Meli, B. Allotta, A. Rindi, A. Capuozzo, and L. Rucher, “Feasibility Analysis of Positioning and Navigation Strategies for Railway and Tramway Applications,” IFAC-PapersOnLine, vol. 53, no. 2, pp. 15680–15686, 2020, doi: 10.1016/j.ifacol.2020.12.2557.

[63]       B. Malakar and B. K. Roy, “Train localization using an adaptive multisensor data fusion technique,” Transport, vol. 34, no. 4, pp. 508–516, 2019, doi: 10.3846/transport.2019.11313.

[64]       B. Allotta, P. D’Adamio, M. Malvezzi, L. Pugi, A. Ridolfi, and G. Vettori, “A localization algorithm for railway vehicles,” Conf. Rec. - IEEE Instrum. Meas. Technol. Conf., vol. 2015-July, pp. 681–686, 2015, doi: 10.1109/I2MTC.2015.7151350.

[65]       G. P. S. Carvalho and R. R. Costa, “Localization of an Autonomous Rail-Guided Robot Using Particle Filter,” IFAC-PapersOnLine, vol. 50, no. 1, pp. 5642–5647, 2017, doi: 10.1016/j.ifacol.2017.08.1112.

[66]       S. Hensel, C. Hasberg, and C. Stiller, “Probabilistic rail vehicle localization with eddy current sensors in topological maps,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 1525–1536, 2011, doi: 10.1109/TITS.2011.2161291.

[67]       F. Boehringer, “Train location based on fusion satellite and train-borne sensor data,” Locat. Serv. Navig. Technol., vol. 5084, no. August 2003, p. 76, 2003, doi: 10.1117/12.487062.

[68]       A. Acharya, S. Sadhu, and T. K. Ghoshal, “Train localization and parting detection using data fusion,” Transp. Res. Part C Emerg. Technol., vol. 19, no. 1, pp. 75–84, 2011, doi: 10.1016/j.trc.2010.03.010.

[69]       R. Shenton, “Train positioning using video Odometry ESGI100 Train positioning using video Odometry Problem presented by.”

[70]       A. Mirabadi, N. Mort, and F. Schmid, “Application of sensor fusion to railway systems,” IEEE Int. Conf. Multisens. Fusion Integr. Intell. Syst., no. January, pp. 185–192, 1996, doi: 10.1109/mfi.1996.572176.

[71]       C. Hasberg, S. Hensel, and C. Stiller, “Simultaneous localization and mapping for path-constrained motion,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 541–552, 2012, doi: 10.1109/TITS.2011.2177522.

[72]       F. Böhringer and A. Geistler, “Comparison between different fusion approaches for train-borne location systems,” IEEE Int. Conf. Multisens. Fusion Integr. Intell. Syst., pp. 267–272, 2006, doi: 10.1109/MFI.2006.265640.

[73]       A. Albanese and L. Marradi, “The RUNE project: the integrity performances of GNSS-based railway user navigation equipment,” pp. 211–218, Feb. 2008, doi: 10.1109/RRCON.2005.186082.

[74]       D. Stein, M. Lauer, and M. Spindler, “An analysis of different sensors for turnout detection for train-borne localization systems,” 2014, doi: 10.2495/CR140691.

[75]       D. Stein, M. Spindler, J. Kuper, and M. Lauer, “Rail detection using lidar sensors,” Int. J. Sustain. Dev. Plan., vol. 11, no. 1, pp. 65–78, 2016, doi: 10.2495/SDP-V11-N1-65-78.

[76]       R. J. Zygowicz, E. A. Beimborn, Z.-R. Peng, and S. Ocatania, “State of the Art in Automatic Vehicle Location Systems,” p. 27, 2008, [Online]. Available: https://www4.uwm.edu/cuts/its/avlapa.pdf.

[77]       A. José, D. Santos, R. Soares, F. Manuel, A. Redondo, and N. B. Carvalho, “Tracking Trains via Radio Frequency Systems,” IEEE Trans. Intell. Transp. Syst., vol. 6, no. 2, 2005, doi: 10.1109/TITS.2005.848369.

[78]       C. C. Bantin, A. Luttgen, C. Schwellnus, and W. Kinio, “Precise positioning of RFID tags using a phased array antenna,” IEEE Antennas Propag. Soc. AP-S Int. Symp., 2012, doi: 10.1109/APS.2012.6349254.

[79]       A. P. Patra and U. Kumar, “Availability analysis of railway track circuits:,” http://dx.doi.org/10.1243/09544097JRRT296, vol. 224, no. 3, pp. 169–177, Feb. 2010, doi: 10.1243/09544097JRRT296.

[80]       P. Goel, S. I. Roumeliotis, and G. S. Sukhatme, “Robust localization using relative and absolute position estimates,” IEEE Int. Conf. Intell. Robot. Syst., vol. 2, pp. 1134–1140, 1999, doi: 10.1109/iros.1999.812832.

[81]       V. OKOROCHA and O. OLAJUGBA, “Comparative Analysis of Short, Medium and Long Baseline Processing inthe Precision of GNSSPositioning.”Engaging the Challenges – Enhancing the Relevance, FIG Congress 2014. Kuala Lumpur, Malaysia.16-21 June 2014.,” no. June, pp. 1–15, 2014.

[82]       N. Li et al., “Indoor and outdoor low-cost seamless integrated navigation system based on the integration of INS/GNSS/LIDAR system,” Remote Sens., vol. 12, no. 19, pp. 1–21, 2020, doi: 10.3390/rs12193271.

[83]       L. Chang, X. Niu, and T. Liu, “Gnss/imu/odo/lidar-slam integrated navigation system using imu/odo pre-integration,” Sensors (Switzerland), vol. 20, no. 17, pp. 1–18, 2020, doi: 10.3390/s20174702.

[84]       S. Kuutti, S. Fallah, K. Katsaros, M. Dianati, F. Mccullough, and A. Mouzakitis, “A Survey of the State-of-the-Art Localization Techniques and Their Potentials for Autonomous Vehicle Applications,” IEEE Internet Things J., vol. 5, no. 2, pp. 829–846, 2018, doi: 10.1109/JIOT.2018.2812300.

[85]       M. Malvezzi, G. Vettori, B. Allotta, L. Pugi, A. Ridolfi, and A. Rindi, “A localization algorithm for railway vehicles based on sensor fusion between tachometers and inertial measurement units,” Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit, vol. 228, no. 4, pp. 431–448, 2014, doi: 10.1177/0954409713481769.

[86]       M. Malvezzi, G. Vettori, B. Allotta, L. Pugi, A. Ridolfi, and A. Rindi, “A localization algorithm for railway vehicles based on sensor fusion between tachometers and inertial measurement units:,” http://dx.doi.org/10.1177/0954409713481769, vol. 228, no. 4, pp. 431–448, Apr. 2013, doi: 10.1177/0954409713481769.

[87]       A. Mirabadi and A. Khodadadi, “Slip and slide detection and compensation for odometery system, using adaptive fuzzy kalman filter,” Sens. Lett., vol. 7, no. 1, pp. 84–90, 2009, doi: 10.1166/sl.2009.1014.

[88]       R. Sharma and R. M. Lourde, “Crosstalk reduction in balise and infill loops in automatic train control,” Proc. 2nd IEEE Int. Work. Adv. Sensors Interfaces, IWASI, no. July 2007, 2007, doi: 10.1109/IWASI.2007.4420020.

[89]       J. Moreno García-Loygorri Antonio Pérez-Yuste César Briso, M. Berbineau Alain Pirovano, and J. Mendizábal, “Communication Technologies for Vehicles,” 2018, Accessed: Dec. 23, 2021. [Online]. Available: http://www.springer.com/series/7411.

[90]       X. Zhan, Z. H. Mu, R. Kumar, and M. Shabaz, “Research on speed sensor fusion of urban rail transit train speed ranging based on deep learning,” Nonlinear Eng., vol. 10, no. 1, pp. 363–373, Jan. 2021, doi: 10.1515/NLENG-2021-0028/MACHINEREADABLECITATION/RIS.

[91]       L. Bazant, A. Toma, and H. Mocek, “Description of methodology for data record sorting and saving Satellite Technology for Advanced Railway Signalling,” no. 687414, p. 114, 2017.

[92]       A. Mirabadi, N. Mort, and F. Schmid, “Design of fault tolerant train navigation systems,” Proc. Am. Control Conf., vol. 1, pp. 104–108, 1999, doi: 10.1109/ACC.1999.782749.

[93]       P. Heide, V. Magori, and R. Schwarte, “Coded 24 GHz Doppler radar sensors: A new approach to high-precision vehicle position and ground-speed sensing in railway and automobile applications,” IEEE NTC 1995 - Microw. Syst. Conf. Conf. Proc., pp. 101–104, 1995, doi: 10.1109/NTCMWS.1995.522870.

[94]       M. Malvezzi, B. Allotta, and M. Rinchi, “Odometric estimation for automatic train protection and control systems,” Veh. Syst. Dyn., vol. 49, no. 5, pp. 723–739, 2011, doi: 10.1080/00423111003721291.

[95]       M. B. Alatise and G. P. Hancke, “A Review on Challenges of Autonomous Mobile Robot and Sensor Fusion Methods,” IEEE Access, vol. 8, pp. 39830–39846, 2020, doi: 10.1109/ACCESS.2020.2975643.

[96]       Y. Liu and Y. Bao, “Review of electromagnetic waves-based distance measurement technologies for remote monitoring of civil engineering structures,” Meas. J. Int. Meas. Confed., vol. 176, no. February, p. 109193, 2021, doi: 10.1016/j.measurement.2021.109193.

[97]       E. Bertran and J. A. Delgado-Penín, “On the use of GPS receivers in railway environments,” IEEE Trans. Veh. Technol., vol. 53, no. 5, pp. 1452–1460, Sep. 2004, doi: 10.1109/TVT.2004.832416.

[98]       “Blog - LiDAR Sensing Applications in the Railroad Industry - LeddarTech.” https://leddartech.com/blog-lidar-sensing-applications-in-the-railroad-industry/ (accessed Mar. 30, 2022).

[99]       T. Daoust, F. Pomerleau, and T. D. Barfoot, “Light at the end of the tunnel: High-speed lidar-based train localization in challenging underground environments,” Proc. - 2016 13th Conf. Comput. Robot Vision, CRV 2016, no. October, pp. 93–100, 2016, doi: 10.1109/CRV.2016.54.

[100]     S. Liu, Q. Wang, and Y. Luo, “A review of applications of visual inspection technology based on image processing in the railway industry,” Transp. Saf. Environ., vol. 1, no. 3, pp. 185–204, 2019, doi: 10.1093/tse/tdz007.

[101]     A. Berg, K. Öfjäll, J. Ahlberg, and M. Felsberg, “Detecting Rails and Obstacles Using a Train-Mounted Thermal Camera,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 9127, pp. 492–503, 2015, doi: 10.1007/978-3-319-19665-7_42.

[102]     T. Ye, Z. Zhang, X. Zhang, and F. Zhou, “Autonomous Railway Traffic Object Detection Using Feature-Enhanced Single-Shot Detector,” IEEE Access, vol. 8, pp. 145182–145193, 2020, doi: 10.1109/ACCESS.2020.3015251.

[103]     D. Ristić-Durrant, M. Franke, and K. Michels, “A Review of Vision-Based On-Board Obstacle Detection and Distance Estimation in Railways,” Sensors 2021, Vol. 21, Page 3452, vol. 21, no. 10, p. 3452, May 2021, doi: 10.3390/S21103452.

[104]     F. Elbahhar and M. Heddebaut, “Advanced Train Positioning/Communication System,” Mod. Railw. Eng., Feb. 2018, doi: 10.5772/INTECHOPEN.71768.

[105]     A. Geistler, “Train location with eddy current sensors,” 2002, Accessed: Jan. 24, 2022. [Online]. Available: www.witpress.com.

[106]     S. Hensel, T. Strausß, and M. Marinov, “Eddy current sensor based velocity and distance estimation in rail vehicles,” IET Sci. Meas. Technol., vol. 9, no. 7, pp. 875–881, Oct. 2015, doi: 10.1049/IET-SMT.2014.0302.

[107]     J. Yin, D. Chen, T. Tang, L. Zhu, and W. Zhu, “Balise arrangement optimization for train station parking via expert knowledge and genetic algorithm,” Appl. Math. Model., vol. 40, no. 19–20, pp. 8513–8529, Oct. 2016, doi: 10.1016/J.APM.2016.04.015.

[108]     B. Malakar and B. K. Roy, “Survey of RFID applications in railway industry,” 1st Int. Conf. Autom. Control. Energy Syst. - 2014, ACES 2014, 2014, doi: 10.1109/ACES.2014.6807999.

[109]     “Track circuit - Wikipedia.” https://en.wikipedia.org/wiki/Track_circuit (accessed Jan. 24, 2022).

[110]     R. F. Brena, A. A. Aguileta, L. A. Trejo, E. Molino-Minero-Re, and O. Mayora, “Choosing the best sensor fusion method: A machine-learning approach,” Sensors (Switzerland), vol. 20, no. 8, Apr. 2020, doi: 10.3390/S20082350.

[111]    Z. Wang, Y. Wu, and Q. Niu, “Multi-Sensor Fusion in Automated Driving: A Survey,” IEEE Access, vol. 8, pp. 2847–2868, 2020, doi: 10.1109/ACCESS.2019.2962554.

[112]     G. Velasco-Hernandez, D. J. Yeong, J. Barry, and J. Walsh, Autonomous Driving Architectures, Perception and Data Fusion: A Review; Autonomous Driving Architectures, Perception and Data Fusion: A Review. 2020.

[113]     S. Jusoh and S. Almajali, “A Systematic Review on Fusion Techniques and Approaches Used in Applications,” doi: 10.1109/ACCESS.2020.2966400.

[114]     F. Castanedo, “A review of data fusion techniques,” Sci. World J., vol. 2013, 2013, doi: 10.1155/2013/704504.

[115]     J. Fayyad, M. A. Jaradat, D. Gruyer, and H. Najjaran, “Deep learning sensor fusion for autonomous vehicle perception and localization: A review,” Sensors (Switzerland), vol. 20, no. 15, pp. 1–34, 2020, doi: 10.3390/s20154220.

[116]     S. Kuutti, R. Bowden, Y. Jin, P. Barber, and S. Fallah, “A Survey of Deep Learning Applications to Autonomous Vehicle Control,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 2, 2021, doi: 10.1109/TITS.2019.2962338.

[117]     D. J. Yeong, G. Velasco-hernandez, J. Barry, and J. Walsh, “Sensor and sensor fusion technology in autonomous vehicles: A review,” Sensors, vol. 21, no. 6, pp. 1–37, 2021, doi: 10.3390/s21062140.

[118]     J. Kim, J. Kim, and J. Cho, “An advanced object classification strategy using YOLO through camera and LiDAR sensor fusion; An advanced object classification strategy using YOLO through camera and LiDAR sensor fusion,” 2019.

[119]     K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition.” [Online]. Available: http://image-net.org/challenges/LSVRC/2015/.

[120]     W. Liu et al., “SSD: Single Shot MultiBox Detector,” doi: 10.1007/978-3-319-46448-0.

[121]     X. Zhou, D. Wang, and P. Krähenbühl, “Objects as Points.” [Online]. Available: https://github.

[122]     D.-J. Jwo and T.-P. Weng, “An Adaptive Sensor Fusion Method with Applications in Integrated Navigation,” 2021, doi: 10.1017/S0373463308004827.

[123]     B. Khaleghi, A. Khamis, F. O. Karray, and S. N. Razavi, “Multisensor data fusion: A review of the state-of-the-art,” Inf. Fusion, vol. 14, no. 1, pp. 28–44, Jan. 2013, doi: 10.1016/J.INFFUS.2011.08.001.

[124]     L. Kong, X. Peng, Y. Chen, P. Wang, and M. Xu, “Multi-sensor measurement and data fusion technology for manufacturing process monitoring: A literature review,” Int. J. Extrem. Manuf., vol. 2, no. 2, 2020, doi: 10.1088/2631-7990/ab7ae6.

[125]     B. Allotta et al., “An innovative localisation algorithm for railway vehicles,” Vehicle System Dynamics, vol. 52, no. 11. pp. 1443–1469, 2014, doi: 10.1080/00423114.2014.943928.

[126]     Z. Wang, G. Yu, B. Zhou, P. Wang, and X. Wu, “A Train Positioning Method Based-On Vision and Millimeter-Wave Radar Data Fusion,” IEEE Trans. Intell. Transp. Syst., pp. 1–11, 2021, doi: 10.1109/TITS.2020.3046497.

[127]     G. Hu and Y. Dong, “A novel static calibration method for antenna position error based on CDGPS,” IET Conf. Publ., vol. 2012, no. 598 CP, pp. 243–247, 2012, doi: 10.1049/cp.2012.0964.

[128]     J. Goya, L. Zamora-Cadenas, S. Arrizabalaga, A. Brazález, J. Meléndez, and J. Mendizabal, “Advanced Train Location Simulator (ATLAS) for developing, testing and validating on-board railway location systems,” Eur. Transp. Res. Rev., vol. 7, no. 3, 2015, doi: 10.1007/s12544-015-0173-5.

[129]     M. Hamadache, S. Dutta, O. Olaby, R. Ambur, E. Stewart, and R. Dixon, “On the fault detection and diagnosis of railway switch and crossing systems: An overview,” Appl. Sci., vol. 9, no. 23, 2019, doi: 10.3390/app9235129.

[130]     “Prediction of Position Fig 5.2 General schematic for mobile robot localization.”

[131]     G. Lederman, S. Chen, J. H. Garrett, J. Kovačević, H. Y. Noh, and J. Bielak, “A data fusion approach for track monitoring from multiple in-service trains,” Mech. Syst. Signal Process., vol. 95, pp. 363–379, Oct. 2017, doi: 10.1016/J.YMSSP.2017.03.023.

[132]     M. Yonekawa and T. Tanaka, “Relations between positioning result and each error factor in GPS,” 2006 SICE-ICASE Int. Jt. Conf., pp. 1361–1365, 2006, doi: 10.1109/SICE.2006.315574.

This entry is adapted from the peer-reviewed paper 10.3390/s22114185