Non-Visual Sensing of Metallic Pavement Markers From a Moving Vehicle

Snow and ice covering paved surfaces cause problems for roads and other vehicles operating in cold regions. There are notable risks for road accidents and reduced performance of vehicle movements, e.g., in industrial facilities such as ports, when road markings and marked pathways are not visually observable. In this study, metallic paved surface markers identified by metal sensors are tested. The sound signals captured by a metal detector attached to the vehicle are used to detect the markings. Results indicate that the tested method for reading markings through snow is effective at lower speeds. The main advantages of the tested technology are effectiveness in snowy and icy surface conditions, low upfront cost detection technology in some environments, robust system without moving parts, high sampling rate, and low operating costs. The procedure is not restricted to specific environmental conditions, such as snow and ice; it also has the potential to detect markings through layers of mud, tree leaves or sand. The detectors have application prospects in intelligent transportation system applications at airports, logistics terminals and industrial facilities.


A. Background and Aim
T HE functioning transport system is an integral component of modern society. It directly contributes to the everyday well-being of its population and the efficiency of industries [1]. However, according to statistics from the European Transport Safety Council, in 2018, more than 25,000 people lost their lives in road accidents in the European Union [2]. Hence, improving road safety is being emphasized in various national and international policies [3]. These policies aim to improve road safety systems through various measures, such as developing safer vehicles, improving the road infrastructure, enhancing post-crash response, developing legislation, and affecting the behavior of road users [4]. It is well recognized that traffic accident risks depend on several interrelated factors, including driver alertness, road environment, type of vehicle, traffic intensity, and speed. Although drivers have long Manuscript  been considered the main cause of road accidents, external factors such as darkness and challenging weather conditions contribute to drivers' errors [5].
Road pavement markings are one piece in this complex road safety puzzle. The markings provide visual guidance for road users and optimize the safe use of road space. They are particularly important in darkness and challenging weather conditions when the amount of light available to the driver's decreases, which narrows and shortens the human field of vision and impairs the perception of color, shape, texture, contrast, and movement [5]. At present, the commonly used materials for road pavement markings are paints, thermoplastics, two-component resins, tapes, and glass beads [3], [6]. Babić et al. found that different road marking effects, such as red median and horizontal warning sign treatments, positively affect driving behavior before and throughout the dangerous curves [7]. Tsyganov et al. presented that roads without edge lines have an 11% higher risk of road accidents than roads with edge lines [8]. The presence of edge lines also had positive effects on road safety in nighttime conditions [5]. National Cooperative Highway Research Program reported that the number of road accidents in night-time conditions decreased by 6% after renewing the road markings, i.e., increasing their retro-reflectivity [5]. Raised pavement markers are a commonly used method to increase road safety [9].
Snow and ice on top of road surfaces not only reduce the capability of drivers to observe road markings but also cause severe problems for traffic in wintertime in general [10], [11]. The magnitude of problems varies depending on the climate, drivers' winter driving experience, and whether vehicles are equipped for winter by, for instance, having winter tires. Strong et al. found that accidents occur 8.7 times more often on snow-covered roads than in ideal conditions, even though average speeds on snowy roads are on average 16% lower [12]. On very icy roads, accidents occur 16 times more often while average speeds are 17% lower [12]. Therefore, finding solutions that aid drivers on snowy and icy roads will most likely have an immediate impact on traffic safety.
In industrial environments, such as airports and maritime ports, moving vehicles need to follow precise pathways to ensure safe and efficient operations. For example, taxiways, aprons, and ground vehicle pathways may be covered by snow and ice [13]. Similar dedicated pathways for ground vehicles can be found in maritime port environments and industrial facilities (movements of forklifts, tow tractors, cranes, etc.) This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ Furthermore, automated vehicles require even more precise position and track control, and these are becoming widespread in advanced container ports, for example [14].
This study focuses primarily on reading road pavement markings from a vehicle at low speeds in winter climatic conditions. Our method is based on metal plates on the road pavement, the audible signals of which are read by a metal detector [15]. Generally, metal detectors are used for finding metal objects underground. Our approach resolves the issue of detecting markings during adverse weather in all seasons. The road pavement marker information based on audio signals is especially important for old drivers who have a longer reaction time than young drivers [16] owing to a gradual loss of visual accuracy, difficulty in close vision, changes in color perception, and problems seeing in challenging weather or night-time conditions [5]. A metal marker could be utilized to identify lane lines. As the vehicle approaches the identified lane line, the vehicle can have left the driving line, which can be observed by the driver employing an audible signal. Consequently, the proposed new concept could also be an enabler for safe driving and operations in challenging conditions. As far as we know from the literature, this is the first time this method is applied to read markings through the snow.

B. State-of-the-Art Road Marking Detection Technology
The ideal marker system on the road pavements should satisfy the following requirements. Firstly, the markers should have day and night detectability (visual, or some other) during adverse weather in all seasons. Secondly, the marker system should be durable and low maintenance. Third, the detection technology should exhibit sufficient accuracy, low cost, high speed, and wide field of view. Several alternative techniques for sensing pavement markings have been suggested. Different types of road markings such as zebra crossings, arrows, stop lines, lane lines and traffic signs can be detected in good weather conditions by visual techniques [17]. Cameras are widely used for lane detection purposes [18], [19], [20]. The advantages of camera techniques include color information, high resolution, and cheap technology [21]. Cameras are sensitive to non-ideal illumination conditions such as darkness, fog, blowing snow, and sun glare [22], [23]. Light detection and ranging technique (lidar) have been used to measure road pavement markers and have the advantages of a large field of view, 3D information, and measurement speed [24], [25], [26]. Jang et al. developed a lidar and camera system for vehicles to detect lane endpoints at approximately 70-80 km/h of highway in both daytime and nighttime [27]. The disadvantages of lidar are high cost, contains moving parts, and limited operation in poor weather conditions (rain and snow) and in strong sunlight [21], [28]. Ma et al. have developed Bayesian multisensor image fusion method for detecting boundary lines on the road [29]. On roads, pavement marking data can be acquired by radio detection and ranging (radar) sensors [22], [28]. Radar sensor makes it possible to see markings in poor visibility conditions at a long distance. In addition, it is small and lightweight. The weakness of the radar sensor is that it is a rather expensive device and images have low accuracy [21]. The automatic steering control of the vehicle has been studied by reading magnetic markers with a radio wave receiver [30], [31], [32]. The advantages of this method are its simplicity and reasonable price. Its disadvantages are that magnet installations are difficult to change, they must be buried on the road surface, and electromagnetic noise [22], [30]. Generally, the vehicle speed is 100 km/h (30 m/s), getting all road markings detected becomes challenging [33].
In sum, the current system for detecting road pavement markers does not meet all ideal requirements (Table I). The most challenging problem is recognizing markers through snow, ice, and dirt [34]. Consequently, a simple and low-cost marker detection method that would allow real-time monitoring in winter climatic conditions is needed. The proposed method requires metal material on the road which is quite an expensive investment. However, investment costs would be low in limited areas such as airports, marine ports, and industrial areas. Generally, the speeds in the urban roads, airports, logistics terminals, and industrial facilities area are below 60 k/km.

A. Metal Detecting Principle
The magnetic flux density B caused by the excitation current is given by Ampere's law (Maxwell's first equitation) [35]. Let us assume that I (t) is the current in the field coil. Biot-Savart's law gives the magnitude of the magnetic field density at the center as follows: where N f is the number of turns in the field coil, μo is the vacuum permeability constant, and Rf is the radius of the field coil [36]. The induced electric field strength E is given by the electromagnetic induction law, known as Faraday's law (Maxwell's second equitation). According to Faraday's law, the electromotive force ε induced in the pick-up coil is = N p d∅ dt (2) where N p is the number of turns in the pick-up coil is the magnetic flux in the pick-up coil due to the magnetic field produced by the field coil. Approximating the magnetic field to be constant throughout the interior of the smaller pick-up coil.
Then the magnetic flux in the pick-up coil can be expressed as where A is area of coin, B is magnetic field, R p is the radius of the pick-up coil. Finally, Eq. (2) can be rewritten as follows: The magnitude of the electromagnetic field detected by the receiver depends on the permeability and conductivity of the medium and the properties of the metal object. Commercial metal detectors use advanced signal processing and phase information to improve detection sensitivity, but the basic principle of physics is still Faraday's law [36]. Fig. 1 shows the diagram for signal processing to identify a metal plate of position. An EQUINOX 800 metal detector (Minelab) equipped with EQX 11 Double-D coil (25 cm diameter) was used. The advantages of a Double-D coil are stability in mineralized ground, good depth, sensitivity, and pinpointing of the target. EQUINOX 800 was chosen for the parameter soil compensation and simultaneous operation of multiple frequencies (5, 10, 15, 20, and 40 kHz). The multiple-frequency method achieves a high level of target accuracy at a depth much better than any single frequency. According to Faraday's law, the magnetic field induces an electromotive force in the metal object. The decay currents generate a secondary magnetic field, and the time rate-ofchange of the field is detected by a receiver coil located at the sensor. The oscillator generates a signal with a frequency and excites the transmitter coil which generates a timevarying electromagnetic field. Phase shift block can be set to prevent false detection or discriminate the specific types of metal. The signal phase shift depends on the conductivity and permeability of an object. The frequency and phase of the reference signal for the synchronous demodulator [37]. The amplified signal is guided into an analog/digital converter of the sound card. The amplified audio signals were recorded at a 44.1 kHz sampling rate on a computer via a sound card (ASUS XONAR U7 7.1 MKII).

B. Measurement System Description
Noise is a possible source of problems for metal detectors to read metal markers on the road. The noise or voltage spike can be caused such as the vibration of coils, instrument noise, temperature fluctuations, power line frequency interference, contamination of coils, and extra small metal objects on the road. The signal to noise ratio improvement can be used by a threshold level that ignores level lower values [38]. Using the smoothing window (Hanning) can be removed the ripple on either side of the audio peak [39]. As a result of smoothing windows, noise and outliers can be removed from the audio signal. In addition, the sensitivity of the metal detector can be adjusted to a suitable value under noise conditions.

C. Testing and Demonstration Set-up
The experiment examined the effect of vehicle speed and snow on the amplitude of the audio signal. The test was conducted at the OuluZone in cold regions in Finland on a spring on the test track of the motorsport center. The metal markers were made of abrasion-resistant steel (Rex 500, SSAB). The metal detector is attached to a stand built on the front of a vehicle (at a distance of 30 cm). The stand allows the metal detector to be moved further away from the front of the vehicle, so it does not receive a signal from the metal parts of the vehicle. The distance of the coil from the ground was 9 cm. The test vehicle was a Toyota RAV. The implementation of the measurement system is shown in Fig. 2.

III. PERFORMANCE TESTING AND EVALUATION
The purpose of the experiment test is to help understand the feasibility and limitations of using metal markings and metal detectors in winter climatic conditions with pavement covered by snow and ice. The performance of the tested system was first carried out by measuring the amplitudes of the audio signal at different ice and snow layer thicknesses. Second, the dependence of the audio signal strength on the distance between the coil and the detector was investigated. Finally, the measurement system was demonstrated to detect pavement markers from a moving vehicle in winter conditions.

A. Performance Testing of the Metal Detector System
The test aims to determine the maximum height of the ice and snow layer that can be detected by the metal detector. Firstly, we deal with the case of the ice layer. For this purpose, we used an abrasion-resistant steel plate with dimensions of  150 (L) × 50 (W) × 5 (H) mm 3 . The steel plate was placed on the bottom of a plastic pail. Ice layers of 3, 6, 12, and 15 cm thickness, respectively, were prepared for the pails with water. The measurement was done by placing a plastic frame (9 cm height) between the ice and the metal detector (Fig.3).
The purpose of the frame is to prevent the metal detector from coming into direct contact with the ice to correspond to the actual situation in the vehicle. The amplitude curves for ice layers as a function of time were measured using the metal detector (Fig. 4). In Fig. 4, the amplitudes are 0.038 for the ice layers under 15 cm thickness, the same as a steel plate without an ice layer. The exception is the background where is no metal on the ground (Fig. 4 a). The higher the sensitivity of the metal detector, the smaller the pieces of metal it can detect, resulting in saturation of the output signal voltage, as shown in Fig. 4 b-e). In principle, it is possible to adjust the sensitivity of the metal detector so that the thickness of the snow or ice layer can also be estimated. In our study, it is important to detect road pavement metal markers through ice or snow.
When the thickness of the ice layer is 15 cm, the amplitude decreases by 53%. However, various parameters also affect the amplitude obtained from the detector in addition to the layer thickness, such as conditions (temperature, moisture, and salt content), marker (size, shape, conductivity, and magnetic susceptibility), and device (size of the coil, and measuring frequency) [40], [41]. Furthermore, Fig. 4 also illustrates the distortion of a sinusoidal waveform caused by harmonics. However, the results indicate the metal detector's ability to read markings through the ice.
In a way similar to that mentioned above, the amplitudes of the audio signal were measured for snow layers that were 3, 6,9,12,15, and 20 cm thick, respectively. The results plotted in Fig. 5 show that the amplitude of signals is the same for all five thicknesses of the snow layer. In other words, the output signal of the metal detector saturates withal snow thicknesses.
The results show that a 20 cm layer of snow gives the maximum amplitude compared to the ice amplitude result (15 cm). According to Finnish Road Administration, Fig. 6. Amplitude signal as a function of the distance between the metal plate and the coil of the metal detector.
the reason for this is that the dielectric value of snow (8)(9)(10)(11)(12) is twice as high as that of ice (4). Based on the results, it could be assumed that the combined effect of ice and snow thickness is minor if the layer thicknesses are less than 12 cm.
For example, based on the instructions of the Finnish Transport Infrastructure Agency (FTIA) [42], the key quality requirement for winter care is that the maximum snow depth is 4 cm for most important roads, such as motorways. Ploughing must be in progress when 2 cm of snow has been accumulated (so-called starting threshold). Compared to the established winter maintenance standard of FTIA, the results show that the device's performance is adequate to find the marker through the snow on the roads.

B. Marker Recognition Test
The purpose of this test is to investigate the distance between the metal plate and the metal detector coil, which can be detected by the metal detector. In the study, we used an abrasion-resistant steel plate with dimensions of 300 (L) × 100 (W) × 5 (H) mm 3 . The distances between the metal plate and the metal detector coil were 0, 5,15,25,35,45, and 55 cm.
The measurement was performed by placing a plastic frame (9 cm height) between the metal plate and the metal detector. The result is plotted in Fig. 6. In the figure, the metal detector found a metal plate at a distance of 45 cm and gave full amplitude already at a distance of 25 cm before the metal plate. This result means that the metal plates should be about 25 cm apart in order to signal is strong. Byun et al. investigated the position of the vehicle based on magnetic markers, which were detected within a distance of 15 cm by the sensor [31].

C. Demonstration of the Detector System on the Test Track
The vehicle's speed was set via a cruise control system at either 10, 30, or 60 km/h through manual adjustment of the vehicle's accelerator pedal. These speeds were chosen as practical speeds in urban areas and other low speed environments. The six steel metal plates (300 (L) × 100 (W) × 5 (H) mm 3 ) were placed at intervals of 6 m on the test road. Thus, interference between the magnetic fields of adjacent metal plates could be ruled out in this study. A snow layer 5 cm thick was added to the metal on top of the plates. The effect of vehicle speed on the amplitude signals for the metal plate and added snow layers is shown in Fig. 7.
The results on the left side of Fig. 7 show that the metal plates (markers) were successfully detected at all speeds without extra peaks. However, the amplitude levels of the signals were not the same, which may be due to the vibration of the detector, the driving line, or the instability of the metal detector (hand-held). The width of the peak amplitude averaged 102 ± 0.15 cm except for the first peak, which was twice as wide. This outcome could be attributed to the automatic gain adjustment of the metal detector. Initially, the sensitivity of the output amplifier was higher without the signal. When the signal comes from the marker, the gain decreases. When the output voltage becomes too high (distortion) and remains so for some time, the width of the amplitude increases.
The results of the snow-covered steel plate, presented on the right-hand side of Fig. 7, suggest that using the metal detector method is effective in extracting road pavement markings through the snow. The metal plates were clearly detected and with adequate accuracy. The metal detector guidance system responds to the special needs of drivers who cannot visually observe the pathway due to poor weather and/or environmental conditions. Our method could read road markings at a speed of 60 km /h under snow.

IV. DISCUSSION
The main contribution of this study was to demonstrate the use of a low-cost metal detector for detecting the metal plate (road marker) under snow from a vehicle moving at urban speed. The method is based on analyzing audio signals from metal plates in the measurement system. There are some conclusions that can be drawn from the demonstration study. First, based on the results, it is possible to identify the markers at a depth of 12 cm and 20 cm under ice and snow. According to the guidelines of the Finnish Transport Agency (FTIA), the maximum allowed depth of snow is 4 cm on motorways [42]. So, the performance of the metal detector system is very sufficient to find a marker through the snow. Furthermore, by adjusting the sensitivity of the metal detector, it is possible to estimate the thickness of the snow or ice layer on the road. Second, the audio signal (lateral) reaches its maximum level at a distance of about 25 cm from the marker, and then gradually weakens farther away from the marker. Third, demonstration results show that the signal generated by metal plates through the snow cover can be read accurately at urban speeds. However, the demonstration results are not restricted to snow cover only but suggest that the proposed method has the potential to detect markings through various types of layers, such as mud, tree leaves, or sand. All these are to be found in defining road conditions in different parts of the world. All these conditions have a negative impact on road safety.
The major advantages of the demonstrated application compared to other techniques include • high sampling rate • no requirements for moving parts • applicability to different road weather conditions • inexpensive, mature detection technology • foreseeably low operating costs (i.e., no internal power source is needed for metal or metalized markers), and • sensitivity to detect markers through different types of pavements covering layers. The metal detector has been installed in the middle of the car bumper; in which case, the metal plates should be placed in the middle of the road lane. Another possibility is that the metal detectors are installed on both ends of the front of the car in which case the middle and the edge line of the road can be read simultaneously. Other metal objects on the road, such as metal caps, also interfere with the measurement to give false detections. Such interference can be prevented by using several metal detectors in the vehicle. The metal detector could complement a camera-based marker detection system to improve reliability in extreme conditions. In other words, metal road markings will supplement visual guidance in darkness, sun glare, snow, and fog as well as in winter conditions. The risk of road accidents is reduced, and path-keeping reliability is enhanced. The technology can be applied to any type of vehicle fleet, such as forklifts, automated shuttles, or maintenance vehicles.
There are some inevitable practical challenges. First, metallic road markings require a major investment if applied to public roads' pavements on a wider scale. The metal plates must be installed on the road at the same time as the road is paved with an asphalt layer. This would ensure that metal markers stay on the lane. These costs must be weighed against road safety benefits, which can be substantial but were not estimated in this study. Investment costs can potentially be reduced by adding metal components to conventional road marking paints. However, if the metal powder is coated with nonconductive paint, the audio signal attenuates increases quickly on the marking paint. Second, metals could also be made into pattern markers on which information from the road could be encoded. This is an important application prospect when road transport is getting increasingly automated. Third, in order to materialize the benefits of road marking reading capability, the automotive fleets would need to be equipped with appropriate metal detectors. Considering private car fleets, the application prospects seem limited and require the automotive industry's engagement with the suggested technology. In addition, there should be at least national renewed standards concerning road markings. Therefore, the prospective use cases can most probably be found in special vehicle fleets in special, dedicated areas where the vehicles operate at low speeds. Such can be found in the context of airports, maritime ports, logistics terminals, and industrial facilities where the installation of both plates and detectors may be straightforward because of t limited scale. For example, in an airport, several special moving equipments needs to follow precise pathways regardless of the pathway surface conditions, e.g., in cases where these are covered with snow, ice, sand, or mud. Any areas with needs of zoning and precise pathway controls, and with special moving vehicles and equipment could make use of the demonstrated technology. The metal recovery speed depends on the operating frequency of the metal detector. Choosing a higher operating frequency enables detection at higher vehicle speeds although this leads to weaker depth detection. Table II summarizes the applicability of the metal detector method to infrastructure applications.

V. CONCLUSION
This paper proposes a new method to detect pavement markers covered by snow with a metal detector on road. Testing results show that the signal generated by metal plates through snow can be read accurately at urban speeds. The major advantages of the demonstrated application compared to other techniques include a high sampling rate, lack of moving parts, independence from weather conditions, and inexpensive detection technology. There are also some practical challenges such as metallic road markings requiring a major investment in road pavements to deploy the technology in the first place. Therefore, the metal detector method is best suited for limited areas such as airports, logistics terminals, and industrial facilities where upfront investment costs are substantially lower and where the vehicles operate at low speeds.
In the future that metal detectors could be an integral part of intelligent vehicle infrastructure (V2I) systems. A higher degree of vehicle autonomy requires markings that are reliably and accurately detected regardless of pavement surface conditions. In addition, the measurement method would be suitable for temporary routes where, for example, autonomous vehicles pass trees from the forest to the road. This capability would pave the way for safer and more reliable autonomous vehicles and thus increase prospects of automation.
ACKNOWLEDGMENT Moreover, the authors thank Jarmo Pekkala of Ni-Ro Oy Oulu, Finland, for his arrangements for the experimental tests at OuluZone and Pasi Virtanen and Krister Wessman of Teknos Ltd., Finland, for their co-operation.