Effectiveness of Active Luminous Lane Markings on Highway at Night: A Driving Simulation Study

: Road lane markings play an essential role in maintaining trafﬁc order and improving trafﬁc safety and efﬁciency. Active luminous lane markings have emerged with advances in technology recently. However, it is still not completely clear what impact their application will have on drivers. This paper aimed to study the effectiveness of active luminous lane markings on highways at night. A driving simulation experiment was carried out based on advanced driving simulators at Tongji University. The driving simulation experiment involved 31 participants and 9 simulation scenes with 6 different types of lane markings models and the same 2-way highway segment, which was 5300-m long with four 3.75-m wide driving lanes. The study participants drove through the simulated highway while the vehicle operation data and the driver’s eyes changing data were continuously captured. Overall, the pupil area change rate, steering wheel speed, brake pedal force, gas pedal, lane departure, and operating speed indicators were selected to evaluate the effectiveness of the active luminous lane markings. The results are shown as follows: (1) the active luminous lane markings have excellent visual recognition performance at night. Compared with the passive luminous lane markings, the active luminous markings can reduce the mental and physical loads of drivers, increase the early braking distance signiﬁcantly, improve the lane-keeping ability and smooth the operating speed; (2) for the speciﬁc parameter settings of the active luminous lane markings at night, the yellow lane markings are better than the white ones, the point-line-type lane markings are superior to the conventional-type ones, and the blinking frequency is reasonable to set, at a moderate level, as 40 times per min. The results suggest that there are positive effects of active luminous lane markings on the promotion of highway trafﬁc safety and efﬁciency at night, providing theoretical support for the popularization and application of active luminous road lane markings.


Introduction
Since the appearance of road lane marking, it has played an essential role in maintaining traffic order and improving traffic safety and efficiency [1,2]. However, the research [3] has pointed out the effectiveness of road lane markings depends on their visibility. If road lane markings cannot be seen and recognized by drivers, their function cannot be fully explored.
The primary way to enhance the visibility of lane markings is to increase their brightness [4][5][6]. According to different light-emitting modes of lane markings, they can be divided into two types: passive luminous and active luminous. As Smadi [7] pointed out, the light-emitting principle of passive luminous lane markings is to reflect other light sources, such as car lights, street lights, and sunlight. This type of lane markings has been widely used because of their retroreflective properties and low cost, approximately

Scenes Design
The high-fidelity driving simulator at Tongji University was used for the experiment. With the same road alignment, the nine scenes in Table 1 were built using SCANeRstudio1.6 software. The 2-way highway segment was 5300 m long with four 3.75-m wide driving lanes, consisting of several straight segments, four curved segments with different radii, and two tunnel segments. The detailed design parameters of the road alignment are shown in Table 2. Six different lane markings in the driving simulation scenes are presented in Table 3. According to the National Standards of China [1], the conventional-type lane marking lines had a pattern of a 6-m stripe and a 9-m gap, and the point-line-type lane markings line had a pattern of a 0.15-m stripe and a 15-m gap. The widths of both types of lane markings line were 0.15 m. In order to eliminate the significant difference between the artificial luminance level in the simulator and the actual luminance level in the real-world condition, the visual recognition distances of the lane marking samples were tested on an actual road before the driving simulation experiment. Taking the visual recognition distance as the control index, the brightness, size, and inclination of lane markings in the driving simulation scenes were adjusted to achieve the same visual recognition distance as reality.          In order to eliminate the significant difference between the artificial luminance level in the simulator and the actual luminance level in the real-world condition, the visual recognition distances of the lane marking samples were tested on an actual road before the driving simulation experiment. Taking the visual recognition distance as the control index, the brightness, size, and inclination of lane markings in the driving simulation scenes were adjusted to achieve the same visual recognition distance as reality.

Participants
The experiment was a repeated measures design. According to the sample size calculation method [30,31], 31 participants (17 males and 14 females) were recruited as the drivers. The age of participants ranged from 22 to 51 years (mean = 37.5; SD = 10.2). All participants held a valid driver's license for more than 3 years. Besides, all of them were physically and mentally healthy with normal eye vision.

Experimental Procedure
Before the formal experiment, the drivers were required to conduct a 10-min preliminary experiment to get familiar with the driving simulator. Then, they completed the nine simulation scenes in turn. To ensure the drivers not to be affected by intrinsic factors (memory function, fatigue, etc.), they were required to have a rest between two successive scenes. Furthermore, each driver completed the driving simulator experiment in random order. The order in which each driver completed scenes 1-9 was random and different. In this way, as far as the whole experiment was concerned, the resulting errors could be offset. During the experiment, the drivers were required to drive in the right lane all the way with the vehicle beams on, using the vehicle high and low beams as their needed. The SCANeRstudio1.6 software can record the drivers' control parameters (steering wheel, brake pedal, gas pedal, etc.) and the vehicle running parameters. The drivers' eyes change parameters were collected through the Dikablis head-mounted eye tracker. Finally, 279 sets of valid test data were available. In order to eliminate the significant difference between the artificial luminance level in the simulator and the actual luminance level in the real-world condition, the visual recognition distances of the lane marking samples were tested on an actual road before the driving simulation experiment. Taking the visual recognition distance as the control index, the brightness, size, and inclination of lane markings in the driving simulation scenes were adjusted to achieve the same visual recognition distance as reality.

Participants
The experiment was a repeated measures design. According to the sample size calculation method [30,31], 31 participants (17 males and 14 females) were recruited as the drivers. The age of participants ranged from 22 to 51 years (mean = 37.5; SD = 10.2). All participants held a valid driver's license for more than 3 years. Besides, all of them were physically and mentally healthy with normal eye vision.

Experimental Procedure
Before the formal experiment, the drivers were required to conduct a 10-min preliminary experiment to get familiar with the driving simulator. Then, they completed the nine simulation scenes in turn. To ensure the drivers not to be affected by intrinsic factors (memory function, fatigue, etc.), they were required to have a rest between two successive scenes. Furthermore, each driver completed the driving simulator experiment in random order. The order in which each driver completed scenes 1-9 was random and different. In this way, as far as the whole experiment was concerned, the resulting errors could be offset. During the experiment, the drivers were required to drive in the right lane all the way with the vehicle beams on, using the vehicle high and low beams as their needed. The SCANeRstudio1.6 software can record the drivers' control parameters (steering wheel, brake pedal, gas pedal, etc.) and the vehicle running parameters. The drivers' eyes change parameters were collected through the Dikablis head-mounted eye tracker. Finally, 279 sets of valid test data were available. 6 Point-line type, yellow, active luminous In order to eliminate the significant difference between the artificial luminance level in the simulator and the actual luminance level in the real-world condition, the visual recognition distances of the lane marking samples were tested on an actual road before the driving simulation experiment. Taking the visual recognition distance as the control index, the brightness, size, and inclination of lane markings in the driving simulation scenes were adjusted to achieve the same visual recognition distance as reality.

Participants
The experiment was a repeated measures design. According to the sample size calculation method [30,31], 31 participants (17 males and 14 females) were recruited as the drivers. The age of participants ranged from 22 to 51 years (mean = 37.5; SD = 10.2). All participants held a valid driver's license for more than 3 years. Besides, all of them were physically and mentally healthy with normal eye vision.

Experimental Procedure
Before the formal experiment, the drivers were required to conduct a 10-min preliminary experiment to get familiar with the driving simulator. Then, they completed the nine simulation scenes in turn. To ensure the drivers not to be affected by intrinsic factors (memory function, fatigue, etc.), they were required to have a rest between two successive scenes. Furthermore, each driver completed the driving simulator experiment in random order. The order in which each driver completed scenes 1-9 was random and different. In this way, as far as the whole experiment was concerned, the resulting errors could be offset. During the experiment, the drivers were required to drive in the right lane all the way with the vehicle beams on, using the vehicle high and low beams as their needed. The SCANeRstudio1.6 software can record the drivers' control parameters (steering wheel, brake pedal, gas pedal, etc.) and the vehicle running parameters. The drivers' eyes change parameters were collected through the Dikablis head-mounted eye tracker. Finally, 279 sets of valid test data were available. In order to eliminate the significant difference between the artificial luminance level in the simulator and the actual luminance level in the real-world condition, the visual recognition distances of the lane marking samples were tested on an actual road before the driving simulation experiment. Taking the visual recognition distance as the control index, the brightness, size, and inclination of lane markings in the driving simulation scenes were adjusted to achieve the same visual recognition distance as reality.

Participants
The experiment was a repeated measures design. According to the sample size calculation method [30,31], 31 participants (17 males and 14 females) were recruited as the drivers. The age of participants ranged from 22 to 51 years (mean = 37.5; SD = 10.2). All participants held a valid driver's license for more than 3 years. Besides, all of them were physically and mentally healthy with normal eye vision.

Experimental Procedure
Before the formal experiment, the drivers were required to conduct a 10-min preliminary experiment to get familiar with the driving simulator. Then, they completed the nine simulation scenes in turn. To ensure the drivers not to be affected by intrinsic factors (memory function, fatigue, etc.), they were required to have a rest between two successive scenes. Furthermore, each driver completed the driving simulator experiment in random order. The order in which each driver completed scenes 1-9 was random and different. In this way, as far as the whole experiment was concerned, the resulting errors could be offset. During the experiment, the drivers were required to drive in the right lane all the way with the vehicle beams on, using the vehicle high and low beams as their needed. The SCANeRstudio1.6 software can record the drivers' control parameters (steering wheel, brake pedal, gas pedal, etc.) and the vehicle running parameters. The drivers' eyes change parameters were collected through the Dikablis head-mounted eye tracker. Finally, 279 sets of valid test data were available.

Participants
The experiment was a repeated measures design. According to the sample size calculation method [30,31], 31 participants (17 males and 14 females) were recruited as the drivers. The age of participants ranged from 22 to 51 years (mean = 37.5; SD = 10.2). All participants held a valid driver's license for more than 3 years. Besides, all of them were physically and mentally healthy with normal eye vision.

Experimental Procedure
Before the formal experiment, the drivers were required to conduct a 10-min preliminary experiment to get familiar with the driving simulator. Then, they completed the nine simulation scenes in turn. To ensure the drivers not to be affected by intrinsic factors (memory function, fatigue, etc.), they were required to have a rest between two successive scenes. Furthermore, each driver completed the driving simulator experiment in random order. The order in which each driver completed scenes 1-9 was random and different. In this way, as far as the whole experiment was concerned, the resulting errors could be offset. During the experiment, the drivers were required to drive in the right lane all the way with the vehicle beams on, using the vehicle high and low beams as their needed. The SCANeRstudio1.6 software can record the drivers' control parameters (steering wheel, brake pedal, gas pedal, etc.) and the vehicle running parameters. The drivers' eyes change parameters were collected through the Dikablis head-mounted eye tracker. Finally, 279 sets of valid test data were available.

Data Analysis
All statistical analyses were conducted using SPSS Statistics software [32]. For the datafitting normal distribution, the analysis of variance (ANOVA) was used for the hypothesis test. This study evaluated the effectiveness of active luminous lane markings from three aspects, i.e., the driver's mental load, physical load, and vehicle running state. Furthermore, six indicators are selected (i.e., pupil area change rate, steering wheel speed, brake pedal force, gas pedal, lane departure, and operating speed).
In the study of Zhang et al. [33], the pupil area change rate was selected as the indicator of the driver's mental load. A greater pupil area change rate was found to be associated with a higher mental load. The mathematical equation for calculating the pupil area change rate is shown as follows: where R i represents the change rate of point i, A i+1 represents the pupil area of i + 1 point, and A i represents the pupil area of point i.
In this experiment, the drivers mainly needed to control the steering wheel, brake pedal, and gas pedal, which represented the drivers' physical loads.
Besides, the positions where the drivers started to brake can be derived from the brake pedal force indicator. When they started to brake, it means that they were aware of the dangerous road ahead. The starting point of each curve and tunnel segments were predefined in the simulation scenes. In this way, the distance between the starting position of braking and the starting point of each curve and tunnel segments can be calculated. Longer distance means that the drivers recognized the adverse road segment ahead earlier.
That is to say, the visual recognition performance of lane markings was better.
The lane departure referred to the distance from the center line of the right lane in this paper. Reflecting the driving stability, it can also be used to judge the consistency of the driver's trajectory and the road alignment. A smaller lane departure means a better road alignment induction effect of lane markings.
The operating speed value can also represent the reflected performance of the lane markings. In terms of the whole driving process, a higher and smoother operating speed means that the lane markings had a better performance for improving traffic efficiency and stability.

Results and Discussion
The mean values, SDs, F statistics, and significance levels of the six selected indicators in the nine scenes were calculated and tested, as shown in Table 4. The results of pupil area change rate are shown in Figure 1.

Pupil Area Change Rate.
The results of pupil area change rate are shown in Figure 1. As shown in Table 4, the pupil area change rates in the nine scenes were significantly different at the 0.01 level. Compared with those in scenes 1 and 2, the change rates of the average pupil areas in scenes 3-9 were smaller. The results showed that the drivers' mental loads in the active luminous lane marking scenes were lower than those in passive luminous lane marking scenes. The average pupil area change rate in scene 2 was smaller than that in scene 1. The average pupil area change rate in scene 4 was smaller than that in scene 3, and the average pupil area change rate in scene 6 was smaller than that in scene 5. The results showed that the yellow lane markings were more conducive to reducing the driver's mental load than the white lane markings at night. The average pupil area change rate in scene 5 was smaller than that in scene 3, and the average pupil area change rate in scene 6 was smaller than that in scene 4, which indicated that the point-line-type active luminous lane markings were more effective to reducing the driver's mental load than the As shown in Table 4, the pupil area change rates in the nine scenes were significantly different at the 0.01 level. Compared with those in scenes 1 and 2, the change rates of the average pupil areas in scenes 3-9 were smaller. The results showed that the drivers' mental loads in the active luminous lane marking scenes were lower than those in passive luminous lane marking scenes. The average pupil area change rate in scene 2 was smaller than that in scene 1. The average pupil area change rate in scene 4 was smaller than that in scene 3, and the average pupil area change rate in scene 6 was smaller than that in scene 5. The results showed that the yellow lane markings were more conducive to reducing the driver's mental load than the white lane markings at night. The average pupil area change rate in scene 5 was smaller than that in scene 3, and the average pupil area change rate in scene 6 was smaller than that in scene 4, which indicated that the point-line-type active luminous lane markings were more effective to reducing the driver's mental load than the conventional-type ones. As for the blinking frequency, the average pupil area change rate in scene 7 was greater than that in scene 6, the average pupil area change rates in scenes 8 and 9 were smaller than in scene 6, and the average pupil area change rate in scene 8 was the smallest, which indicated that a moderate blinking frequency (40 times per min) showed more positive effects on the reduction of the drivers' mental load at night.

Driver's Physical Load
As shown in Table 4, at the 0.01 level, the population means of the brake pedal force and gas pedal in the nine scenes were significantly different, while those of the steering wheel speeds in the nine scenes were not significantly different. In order to integrate these three indicators, the Min-Max method was used to normalize them calculated by the following equation: where x represents the normalized value, and x min and x max represent the minimum and maximum values of this indicator, respectively. According to Equation (2), the steering wheel speed, brake pedal force, and gas pedal were normalized, and the results are shown in Table 5. It was found that when the drivers judged that the road ahead was dangerous or they were in a panic state, they operated the steering wheel faster or step on the brake pedal more forcefully. On the contrary, when the drivers judged that the road ahead was safe, they controlled the steering wheel smoothly and press the gas pedal to get a faster speed. In this experiment, the design of road alignment was gentle. Therefore, a faster steering wheel speed and a greater brake pedal force implied a greater difference between the driver's cognition of the road alignment ahead and the road design itself, that is to say, the lane markings had a worse alignment guidance effect. The lane markings had a better linear guidance effect with a larger gas pedal and a smaller difference between the driver's cognition of the road ahead and the road design itself. Therefore, when calculating the driver's physical load, the steering wheel speed and brake pedal force indicators were negative, while the gas pedal was positive, shown as follows: where L p represents the driver's physical load, N s represents the normalized value of the steering wheel speed, N b represents the normalized value of the brake pedal force, and N g represents the normalized value of the gas pedal. The calculated results are shown in Table 5, and a larger value means a better effect of lane markings. According to the same comparison method as the pupil area change rate in Section 3.1, we can draw the following conclusions. Overall, the driver's physical loads in the active luminous lane marking scenes were lower than those in the passive luminous lane marking scenes. This result is consistent with the research conclusion of Horberry et al. [34], which found that the workload was rated as lower for the enhanced markings. As for the color and type, the yellow lane markings were more effective in reducing the driver's physical load than the white ones at night. The point-line-type active luminous lane markings were more conducive to reducing the driver's physical load than the conventional-type ones at night. Different from the conclusion drawn by the pupil change rate, in terms of the driver's physical load, a higher blinking frequency (60 times per min) showed more positive effects on the reduction of the driver's physical load at night.

Brake Distance
We found that the drivers decelerated before entering the curve and tunnel segments. As analyzed in Section 2.4, the visual recognition distance of the lane markings can be obtained from the brake pedal force. The distances between the point where the drivers started to brake and the starting point of next segment were calculated in Table 6. A larger visual recognition distance means that the drivers braked earlier. For the average visual recognition distance of the lane markings in Table 6, similar to the study of Bella et al. [35], we found that the visual recognition distances of the active luminous lane markings were greater than those of the passive luminous lane markings. For the color and type of the active luminous lane markings at night, yellow was better than white, and the point-line type was better than the conventional type. Consistent with the conclusion drawn by the pupil change rate, from the perspective of the visual recognition distance, a moderate blinking frequency (40 times per min) was more effective for improving the visual recognition distance at night.

Lane Departure
The lane departures of the nine simulation scenes are shown in Figure 2, with the values greater than zero means biased to the left and those less than zero means biased to the right. As shown in Table 4, the lane departures of the nine scenes were significantly different at the 0.01 level. By comparing lane departures of the nine scenes, similar conclusions can be found with the pupil change area rate. Furthermore, the areas of the shaded portion in Figure 2 were calculated to evaluate the lane departure quantitatively. Three areas, i.e., areas above and below the zero line and the total area, were calculated separately, as shown in Table 7.   Table 4, the lane departures of the nine scenes were significantly different at the 0.01 level. By comparing lane departures of the nine scenes, similar conclusions can be found with the pupil change area rate. Furthermore, the areas of the shaded portion in Figure 2 were calculated to evaluate the lane departure quantitatively. Three areas, i.e., areas above and below the zero line and the total area, were calculated separately, as shown in Table 7.   [36] found the effectiveness of passive luminous longitudinal edgeline pavement markings. Liu et al. [37] confirmed the conventional passive luminous markings satisfy spatial lane length requirements by increasing the lanekeeping ability. Based on the above analysis, it was found that the lane marking in scene 8 was the best in this paper. The lane-keeping ability in scene 8 was improved by 31% compared with that in scene 1.

Operating Speed
The operating speeds of the nine simulation scenes are shown in Figure 3. On the whole, the operating speeds in the active luminous lane markings scenes were smoother than those in the passive luminous lane markings scenes. Meanwhile, it was found that the active luminous lane markings increased the driving speed by comparing the means of the operating speed. The smoother and higher operating speeds are possibly associated with the lower mental and physical loads in the active luminous lane marking scenes, leading to faster but more controlled driving. This result is consistent with the study of Ranney and Gawron [38]. Therefore, it can be concluded that the active luminous and yellow lane markings are more effective to increase driver confidence and improve running efficiency than the passive luminous and white lane markings at night. However, for the type and blinking frequency, the conclusions are different from the pupil area change rate. Since the means of the operating speeds in scenes 5 and 6 were smaller than those in scenes and yellow lane markings are more effective to increase driver confidence and improve running efficiency than the passive luminous and white lane markings at night. However, for the type and blinking frequency, the conclusions are different from the pupil area change rate. Since the means of the operating speeds in scenes 5 and 6 were smaller than those in scenes 3 and 4, we cannot accurately conclude that point-line-type active luminous lane markings are better than conventional-type ones at night. From the perspective of the operating speed, a low blinking frequency (20 times per minute) showed more positive effects on the improvement of running efficiency at night.

The Overall Comprehensive Effectiveness Evaluation
The above content evaluated the effectiveness of the active luminous lane markings from the perspective of a single indicator. In this section, a comprehensive evaluation method on the effectiveness of the active luminous lane markings is given. According to Equation (2), the six indicators in Table 4 were normalized, and the results are shown in Table 8.

The Overall Comprehensive Effectiveness Evaluation
The above content evaluated the effectiveness of the active luminous lane markings from the perspective of a single indicator. In this section, a comprehensive evaluation method on the effectiveness of the active luminous lane markings is given. According to Equation (2), the six indicators in Table 4 were normalized, and the results are shown in Table 8. The above six indicators completely included the driver's mental, physical, and vehicle running status. The effectiveness of the lane markings can be evaluated through an integrated indicator E lm , a newly proposed indicator that evaluated the effectiveness of the lane markings in this paper. As analyzed in Section 2.4, for the two indicators-gas pedal and operating speed, a larger value means a better effect of the lane marking, so these two indicators were set as positive values. For the other four indicators, a smaller value means a better effect of the lane markings, so these four indicators were set as negative values. The calculation equation of E lm was written as follows: where E lm represents the comprehensive effectiveness of the lane markings, N p represents the normalized value of the pupil area change rate, N s represents the normalized value of the steering wheel speed, N b represents the normalized value of the brake pedal force, N g represents the normalized value of the gas pedal, N l represents the normalized value of the lane departure, and N o represents the normalized value of the operating speed. The calculation results of E lm are shown in Table 8. Based on the E lm of the nine lane markings scenes, some conclusions can be drawn. Compared with those in scenes 1 and 2, the E lm in scenes 3-9 were greater. It was shown that the active luminous lane markings were comprehensively effective compared with the passive luminous lane markings at night. In terms of the color, the E lm in scene 4 was greater than that in scene 3, and the E lm in scene 6 was greater than that in scene 5. It was shown that the yellow active luminous lane markings were more effective than the white active luminous lane markings at night. For the types, the E lm in scene 5 was greater than that in scene 3, and the E lm in scene 6 was greater than that in scene 4, which indicated that the point-line-type active luminous lane markings were more effective than the conventional-type active luminous lane markings at night. As for the blinking frequency, the E lm in scene 7 was less than that in scene 6, the E lm in scenes 8 and 9 were greater than that in scene 6, and the E lm in scene 8 was the maximum, which showed that a moderate blinking frequency (40 times per min) was more effective at night. The comprehensive effectiveness of the active luminous lane markings is beneficial to promote traffic safety on highways at night.

Application Prospect Analysis
Active luminous lane markings have excellent visual recognition performance at night. They can be used in high-risk road segments caused by poor road alignment conditions such as small radii, long longitudinal slopes, bridges, tunnels, and complicated meteorological conditions such as rain, fog, and ice. Furthermore, active luminous lane markings could be improved. Not only can they emit light actively, they can also have some informational and digital functions. At the same time, a rescue management platform could be built based on intelligent active luminous lane markings. The platform can control and reduce secondary accidents caused by the sudden release of meteorological and unexpected events.
The point-line-type active luminous pavement markings are buried in the asphalt concrete pavement. In practical use, there are three power supply modes: one is to build a solar-power-generation panel on the surface of a block to generate electricity by using solar energy; the second mode is to embed wires under a block to supply power to all blocks through wired transmission; the final mode is to build a radio receiving module inside a block and use a mobile vehicle with a wireless charging module to charge blocks on the road regularly. None of these three methods will significantly increase the cost of infrastructure construction.
Intelligent active luminous lane markings have the following potential advantages. Firstly, intelligent active luminous lane markings can be used for forwarding road alignment guiding and vehicle wake display. According to the real-time detection of meteorological conditions, vehicle type, vehicle speed, events, and other traffic parameters, the traffic risk can be divided into different levels, and the corresponding traffic control strategy is developed. The color of markings can be changed in real-time, such as green for normal traffic, yellow for warning, and red for prohibition. It can not only improve the visual distance but also deliver information, which is helpful to promote traffic safety. Besides, by setting a certain blinking frequency and switching different color schemes of lane markings, it will also produce good results for preventing fatigue driving. Compared with the use of directional rumble strips proposed by Xue et al. [39] to promote driving safety, the intelligent active luminous lane markings proposed in this paper will not cause vibrations to vehicles, thereby making drivers and passengers more comfortable.
Secondly, many other kinds of innovations could be carried out in active luminous lane markings. For example, light-transmitting concrete could be introduced into highways.
Then, intelligent active luminous lane markings can be composed of light-transmitting concrete and LED [40]. This kind of lane markings is embedded in the pavement. Besides, intelligent active luminous lane markings are tightly integrated with new pavement structures such as piezoelectric pavement and solar photovoltaic pavement [41]. They adopt an integrated power supply mode of piezoelectric and solar energy. This is a fusion practice of green highway, safe highway, and smart highway concept.
Thirdly, the identification of lane markings in all weather conditions is a technical problem restricting autonomous vehicles. Similar to the idea of Mao et al. [42] that applies markings in the dynamic reversible lane of an intelligent cooperative vehicle infrastructure system, based on the intelligent active luminous lane markings imagined in this paper, electronic communication components could be embedded to improve the digitization and intelligence levels of the lane markings. In this way, vehicle-to-road coordination can be realized. This provides a new solution for the identification of lane markings in complex weather conditions for autonomous vehicles.

Conclusions
This paper studied the effectiveness of active luminous lane markings on highways at night based on driving simulators. The main conclusions are summarized as follows:

•
The active luminous lane markings had a better visual recognition performance at night, which was improved by about two times compared to the passive luminous lane markings.

•
Compared with the passive luminous lane markings, the active luminous lane markings can reduce the mental and physical loads of drivers, increase the early braking distance by approximately two times, improve the lane-keeping ability by approximately 31%, smoothing the operating speed and be more conducive to improving traffic efficiency and stability. These improvements have great potential for enhancing road safety in night conditions. • For the specific parameter settings of the active luminous lane markings on highways at night, the yellow lane markings were better than the white ones, the point-line-type lane markings were superior to those of the conventional-type ones, and the blinking frequency was reasonable to set, at a moderate level, as 40 times per min.
These findings have theoretical significance for the popularization and application of active luminous road lane markings. Even so, we still have some work to do in the future. Next, we will further study the control strategy of intelligent active luminous lane markings. Different control methods should be studied according to different levels of traffic risk.