Drivers’ Eye Movement Characteristics in a Combined Bridge-Tunnel Scenario on a Mountainous Urban Expressway: A Realistic Study

. Combined bridge-tunnel scenarios of driving on mountainous city expressways occur when bridges and tunnels frequently alternate during driving. Te complex nature of these driving scenarios imposes crucial requirements on the drivers’ eye movement characteristics. Tis paper attempts to clarify these characteristics using descriptive statistics and the box graph method, registering the pupil diameter, blink duration, fxation, saccades, and fxation loci at diferent tunnel locations, bridges, and ramps. Realistic driving experiments were performed on the road segment spanning from the Nanchang tunnel to the Liujiatai tunnel freeway in Chongqing, China. Eye movement data were collected for 21 drivers. Te experimental results showed that, while driving in the tunnel, the maximal pupil diameter of the participating drivers was approximately 4.0mm as the driving mileage and the number of tunnels increased, and the maximal visual load on the drivers in the tunnel tended to be stable. At the second tunnel exit, the ramp, the middle section of the frst bridge, and the third tunnel exit, the driving load was the highest, while the fxation duration was shorter for nighttime driving. Te fxation duration was the longest for the diversion road of bridge B1 to the ramp during the day, and the fxation times were the longest at the beginning and end of the test road. Te drivers more often paid attention to the speed dashboard while entering tunnels during daytime driving (compared with nighttime driving).


Introduction
Te special geographical environment and geological conditions of mountainous cities necessitate the use of tunneltype and elevated expressways on cross-river bridges between diferent city areas to shorten the driving distance and improve the road networks' efciency, given that the trafc fows of multiple ordinary city roads converge to bridges and/or tunnels. With the increased construction of bridgetunnel expressways connecting diferent city areas, defects associated with underground roads and surface trafc connection sections continue to highlight the signifcance of cross-river trafc pressure, demonstrating a "supply exceeds demand" situation [1]. Interchange spacing on mountainous city roads is narrow, and with many openings along the route, interchanges may become very complex [2]. Te visual environment at the typical tunnel entrance and exit also may change signifcantly, which may induce temporary ocular blindness [3]. Bridges and interchanges are often connected by ramps for diversion purposes. When driving on bridges, it is necessary to choose driving lanes in advance according to the guidance provided by trafc signs, and an overload or a lack of relevant information when approaching a diversion ramp is likely to increase the drivers' recognition load and prevent from quickly detecting and processing driving-related information. In turn, this may lead to delayed or incorrect judgment of the drivers and may afect their driving decisions (such as those related to the speed control and lane changes in critical sections or making safe and timely lane changes), which may result in vehicle detours and reduced trafc efciency [4]. As key components of any trafc network, bridges have limited length and are often connected to interchanges, which can easily afect downstream trafc efciency in the case of congestions or accidents. Te rescue time is long and difcult, which may seriously afect the driving safety [5]. On those road sections featuring both bridges and tunnels, congestions at the entrance to and exit from the tunnel are likely to afect the bridge trafc as well, resulting in reduced trafc efciency, increased risk of accidents, and psychological pressure on drivers [6].
In a study on the eye movement characteristics and safe behavior of drivers on tunnel-featuring road sections, it was found that the drivers' visual responses were slower in monotonous driving environments [7]. Te complexity of the driving environment may increase the physiological load on the driver. Te ratio of the pupil diameter change, maximal instantaneous velocity of the pupil area, and duration of converted visual concussion are usually used for quantifying the drivers' state of adaptation to bright illumination [8]. Drivers are more aware of their surroundings when driving in tunnels at high altitudes, and the visual fxation is longer when driving in tunnels. When driving on plain areas, more attention is paid to the road ahead [9]. When a driver enters a super-long tunnel road section, the pupil area changes rapidly, and the driver's psychological state tends to stabilize and relax gradually [10]. Te brightness level of the tunnel entrances and exits afects driving behavior and safety [11]. Te tunnel walls can be decorated to modulate the driving environment-related information and improve the drivers' driving [12]. Te longer the reaction time of a driver, the higher is the likelihood of an operational error, which in turn may lead to safety problems [13]. Te entrance and exit of a typical tunnel are often asymmetrical, and diferent tunnel sections may exert diferent efects on drivers [14]. Driving in tunnels is more risk-prone; thus, there is an urgent need to develop strategies and actions for improving the drivers' awareness of tunnels' driving safety [15].
Among the diferent tunnel groups, the connecting areas exhibit the highest rate of collisions, followed by the entrance into the tunnel. At nighttime, the rate of accidents occurring in tunnels is lower than that in the other tunnel sections [16]. Currently, the recommended safety length threshold for the tunnel connection area is 100 m [17]. Te level of visual adaptation at the upstream tunnel exit and the length of the connection are the main factors afecting the visual adaptation of drivers in the downstream sections of the tunnel and the entire tunnel group section [18].
A study on the cross-river bridge and tunnel connection sections of mountainous city roads found that the likelihood of drivers' repeated fxation is higher in tunnels, while their visual search efciency is lower when driving in tunnels compared with driving on out-of-tunnel road sections. Drivers receive information mainly from straight-up front and straight-down front regions, and the likelihood of fxation in these two areas of the threshold and exit sections is signifcantly higher than that for other road sections [19]. When passing through a tunnel's entrance and exit, the rate of the increase in the pupil diameter varied signifcantly across the tested cohort [20]. As the driving time increased, the blink frequency increased, the blink duration increased, and the visual comfort of driving decreased [21,22].
Many studies have addressed combined bridge-tunnel scenarios of mountain highways, but only a few studies have been conducted on the eye movement characteristics and operational safety of drivers for combined bridge-tunnel mountainous city driving scenarios. Although mountain city expressways (of relevance to combined bridge-tunnel driving scenarios) and mountain highways are both closed fast-driving environments, there are signifcant diferences between them in terms of their external environments, structures and facilities, trafc composition, and trafc operation characteristics. Te structure of a typical combined bridge-tunnel scene frequently changes, with rapid and repeated illumination changes experienced by drivers, and requires the drivers to exert more visual efort while driving on these roads. Te eye movement characteristics in combined bridge-tunnel driving scenarios on mountainous city expressways remain elusive.
Terefore, this study considered a typical mountainous city combined bridge-tunnel scenario, spanning from Nanchang tunnel to Liujiatai tunnel in Chongqing City, for real-time vehicle testing. Eye movement data were collected in real time for test drivers. Tis study elucidates the distribution characteristics of the pupil diameter of the drivers in combined bridge-tunnel driving scenarios and changes in the blink, fxation, and saccade patterns, providing data support and a theoretical basis for tunnel lighting layout and trafc sign setting in mountainous city combined bridge-tunnel scenarios.

Participants.
It is crucial to select the experimenters to increase the reliability and universality of the experimental results. Te selection indicators involved the number, age, occupation, and gender of experimenters. Before determining the number of experimenters, the required sample size needs to be calculated by a reasonable statistical method using the expected variance, target confdence, and margin of error [23], as shown in the following equation: Here, n is the sample size; Z is the standard normal distribution statistics; σ is the standard deviation; E is the maximum error.
A signifcance level of 10% is chosen to refect a 90% confdence level regarding the unknown parameter [24,25]. When the confdence level is 90%, Z is equal to 1.25. σ ranges between 0.25 and 0.5. Due to the infuence of trafc fow and test time on the test section, the value of σ is set to 0.36. E is equal to 10%, and thus, the required sample is 21. Terefore, the sample size is in line with the requirement. In order to statistically determine whether the number of subjects is sufcient for this study, twenty-one drivers were selected as naturalistic driving experiment participants, consisting of 16 males (76%) and 5 females (24%), and each subject had a valid Chinese driver's license with more than 2 years of driving experience, and the age range was from 24 to 45 years old (mean � 32.8; SD � 6.1) in this study. All subjects held a license and had 3-22 years of driving experience (mean � 9.9; SD � 4.6) who drove at least 6,000 km. All participants held a valid driver license, corrected visual acuity above 1.0, normal color vision, and stereovision, without refractive error, amblyopia, strabismus, or other ophthalmic diseases. Every participant had a good willingness to participate. Te experiment complied with the ethical principles of the Helsinki Oath [26,27]. SPSS 25.0 is used to perform a one sample t-test efcacy analysis. Te results show that at the 90% confdence level, the statistical power of the sample size is 0.836, which is > 0.80. Ten, gender was divided into 2 groups (male � 1, female � 2), age into 4 groups (0 for ≤25 years old, 1 for 25∼35 years old, 2 for 35∼45 years old, and 3 for ≥45 years old), the driving years into 3 groups (0 for ≤5 years, 1 for 5∼10 years, and 2 for ≥10 years), and the mileage into 4 groups (0 for ≤100,000 km, 1 for 100,000∼300,000 km, 2 for 300,000∼500,000 km, and 3 for ≥500,000 km), the occupation is divided into 3 groups (0 for employees of state-owned enterprises and public institutions, 1 for professional driver, and 2 for freelancer). Te independent sample Mann-Whitney U test and independent sample Kruskal-Wallis test were used to analyze the infuence of fve factors, such as gender, age, driving age, accumulated driving miles, and occupation, on the 'driver's pupil diameter in the starting point of the test. Te progressive signifcance (2-sided test) was 0.313, 0.335, 0.484, 0.870, and 0.952, respectively, all greater than 0.05, indicating that the original hypothesis was valid, i.e., the test drivers' gender, age, driving years, mileage, and occupation had a signifcant efect on the pupil diameter of the drivers at the starting point of the test. Te grouping of the fve factors did not have a signifcant efect on the pupil diameter of drivers in the starting section of the test. Terefore, the sample size for this experiment can provide reliable answers to the studied questions.

Apparatus.
Te experimental equipment includes three parts, i.e., a dashcam, an eye tracker, and electric vehicles. Te dashcam can record the driving process information with relatively high accuracy. Te information is more comprehensive, ensuring the consistency between the time and location of the road section in the posttest. Tis can help determine the driving location and driving environment [28]. Furthermore, human mental activity is crucial and related to the spatial-temporal characteristics of eye movements, which are extracted by the eye tracker [29]. Te eye tracker is an ultralight and robust noninvasive headtracking module that ensures driver comfort and freedom of behavior [30]. Te parameters, which are usually collected by eye trackers, include the fxation point, trajectory diagram, eye movement time, saccade direction, pupil size, and blink ( Figure 1).

Experimental Road.
In this study, from the Nanchang tunnel to the Liujiatai tunnel in Chongqing was selected as the experimental road. Te length of the experimental road is 8537 m, which is a tunnel-bridge-ramp connecting road. According to the Code for Urban Road Route Design (CJJ193-2012, China) [31], the Nanchang tunnel is 1,164 m long with a cross-sectional width of 10.5 m, a net height of 5 m at the building boundary, and 4-lane road. Te main part of the Caiyuanba Yangtze River bridge is 800 m long, which is the main trafc road that connects Nan-An district and Yu-Zhong district. Te upper deck of the bridge is a 2-way 6lane urban expressway with a design speed of 60 km/h, and the lower deck of the bridge is a double-track urban railway of line 3 with a design speed of 75 km/h. Te bridge and tunnel in Zeng-Jiayan has a total length of 5.54 km, with a designed speed of 60 km/h, which is the main trafc road that connects Jiang-Bei district and Yu-Zhong district. Te standard width of the Zeng-Jiayan Bridge is 32.6 m, including 2 m sidewalk, 1.8 m stifening stringers, 0.5 m crash barriers, 11 m carriageway on the left and right sides of the road, and a 2 m central divider in the middle of the road.
Te bridge-to-tunnel ratio refers to the proportion of the mileage of bridges and tunnels to the total mileage, which is generally used in line engineering (railway, highways, and pipeline); a greater ratio of bridge-to-tunnel means a more difcult project. Te ratio of bridge-to-tunnel of the experimental road is 88%.
Te experimental road was classifed into 6 parts according to type and quantity. Te Nanchang Tunnel, Zeng-Jiayan Tunnel, and Liujiatai Tunnel are coded as T1, T2, and T3, respectively; the Caiyuanba Yangtze River Bridge and Zeng-Jiayan Bridge are coded as B1 and B2, respectively; and the Caiyuanba Ramp is coded as R.
Te alignment and driving-related environment of the studied experimental road are shown in Figure 2  Journal of Advanced Transportation 3 s design speed travel inside and outside the urban underground road opening should be consistent; in difcult conditions, safety measures should be taken [33]. Te design speed for the test section was 60 km/h, and the alignment of the tunnel entrance/exit within 50 m was calculated to meet the consistency requirements. Terefore, a detailed analysis of the eye movement characteristics 50 m before and after the tunnel entrance/exit was carried out in this study. Considering that the entrance into T1 tunnel is a signalized intersection followed by 6% of the underpass road entering the tunnel, drivers are signifcantly disturbed when driving, and the 50 m long road segment before the entrance into T1 tunnel was not included in the present analysis; only the 50 m long road segment immediately following the tunnel entrance was analyzed. Te specifc information is listed in Table 1.

Procedure.
Before the experiment, check whether the experimental equipment such as the eye tracker is normal to avoid failure afecting the process during the experiment. Te Tobii eye tracker was connected to the computer, and the visual calibration of the experimenters was performed to ensure that all parameters could be efectively collected. Make each driver perform a pretest, i.e., become familiar with the test route and test equipment. At the beginning of the formal experiment, the drivers followed their normal driving habits. Te recording personnel pay close attention to the data collection of the recording platform of the eye tracker and remind and adjust the experimental equipment in time to ensure the data quality of the simulation experiment if the data collection is delayed or lost. After the experiment, sort and save the experimental data of each experimental driver. Figure 3 is the experimental fowchart.

Results and Discussion
Eye movement experiments are based on human visual analysis to infer human information processing and psychological cognitive processes. Human eye movements generally have three forms, such as fxation, blink, and saccades. Te pupil diameter indicates the visual adaptability and load of drivers. With a shorter blink duration, the participants had to pay more attention to a more difcult task. Stern et al.believed that the longest blink duration was 500 ms, while there was no consistent conclusion about the shortest blink duration [34]. Benedetto [36]. Fixation refers to the eye movement behavior that stays on the target object for more than 100 ms, the average fxation time represents the time it takes for a driver to process potentially dangerous information, or the ease of extracting valid information.
Usually, the density of information in the fxation area and the ease of processing information directly afect the average fxation time. Saccades occur between two fxations of the human eye, and saccades are defned as the eye staying on the target information for less than 100 ms. Tis indicator describes the process in which the driver searches for target information in the trafc environment [37].
In this study, we selected the eye movement indicators, including the pupil diameter, number of blinks, average blink time, number of fxations, average fxation time, total fxation time, number of saccades, and average saccade time, for quantitative analysis. Explore the inner psychological changes of the tested drivers when driving on the bridge and tunnel combination scenarios of mountainous urban expressways as shown in Table 2.
Due to the disturbance of driver activity and the occlusion of eye images during the experiment, there were problems of eye movement point loss and data anomalies, which were compensated by linear interpolation of signal packet loss, replacement of anomalous values, and noise reduction by a sliding mean flter. Ensure the quality of valid data, invalid data, such as blank data, duration greater than 75 ms, eliminate fxation durations less than 50 ms, and other failure data. Use the method of linear interpolation to compensate for experimental data with an acquisition blanking duration below 75 ms. Use sliding mean fltering and sliding root mean square fltering to reduce the noise of eye movement data.

Descriptive Statistics of Pupil Diameter.
Te eye movement data of real drivers during the daytime fat peak period and free fow at night were collected through a real vehicle experiment. Te pupil diameter test and descriptive statistical analysis were conducted for 6 structures, including T1-T3, according to the division of entrance, middle section, and exit, as shown in Table 3.
Changes in the pupil diameter at the tunnel entrance and exit sections are mainly caused by the changing luminance and involuntary physiological responses that are evoked for adapting to this switch in the drivers' external environment; consequently, pupil diameter changes have been used for characterizing the extent of the drivers' visual load. Under normal conditions, the mean pupil diameter of humans is 2-4 mm during the day and 5-7 mm at night [38]. Using the Origin software to plot the drivers' pupil diameter change curve, we determined that the average pupil diameter of daytime drivers decreased from 4.84 mm to 4.10 mm for the middle section of the tunnel, while the mean pupil diameter of nighttime drivers gradually decreased from 4.71 mm to 3.96 mm, as shown in Figures 4 and 5. Te maximal pupil diameter gradually decreased with increasing driving mileage and number of tunnels, while the maximal visual load for driving in the tunnel gradually stabilized. During the day, the pupil diameter of the drivers increased rapidly when entering the tunnel and decreased linearly when leaving the tunnel. Te pupil diameter was at a normal level for the other road sections. At night, the diference in the pupil diameters of the drivers between the road sections was not obvious, and   the visual load was high, which was afected by the lighting environment. Overall, the pupil diameter was more stable at night than during the day because the lighting of bridges and tunnels was similar at night, while there is a great diference between daytime tunnel lighting and sunlight on other roadways. Te pupil diameter was the largest and signaled a higher load for driving at nighttime on the ramp, which was owing to poor lighting conditions on both sides of the on-ramp and the inability to see the lane lines. Te drivers relied only on the vehicle lights to ensure they drove in the right direction. To improve the safety of nighttime on-ramp driving, it is suggested to install refective flms on both sides of on-ramps and to enhance the recognition ability of drivers. Good lighting of a combined bridge-tunnel scenario at night can relieve the driving load of drivers.
When driving in the tunnel, the average pupil diameter was exhibited an "increasing-decreasing" trend at the entrance, middle, and exit road sections during daytime driving. However, the distribution of the average pupil diameter was more concentrated, without obvious regularity, for nighttime driving, as shown in Figure 6.

Blink Behavior Analysis.
Increasing the difculty of driving tasks would decrease the blink frequency and blink duration [39]. Te blinking duration is a sensitive and reliable index to determine the drivers' visual load, and the frequency of blinking duration distribution of drivers under high mental load was 70-100 ms [40]. Figure 7(a) shows that the average daytime blink duration exceeded 100 ms for the road sections T2-3, R-1, and B1-2, while Figure 7(b) shows that the average nighttime blink duration was the longest for the road sections R-1, B2-1, and T3-3, that is, the drivers' driving load was higher for the second tunnel exit, ramp, middle of the frst bridge, and third tunnel exit. Tis was because, when driving on the frst bridge, the drivers' needed to select the driving lane ahead of time according to the trafc signs, and the guidance-related information before the diversion ramp was overloaded or lacking, increasing the load on the drivers. After identifying the correct passage onto the ramp, vehicles had to split and merge with trafc in the other directions on the ramp section. With increasing ramp trafc, gradually forming intertwined congestion at the diversion point, the shorter length of the intertwined area often created a trafc bottleneck point, and frequent lane changes and braking behavior increased the driving load. Te second tunnel exit was at a short distance after the river bridge, and when driving in the tunnel, the middle space of the city tunnel was relatively closed, the sidewalls were mostly low-contrast white tiles, and the driving landscape was monotonous. Consequently,  Journal of Advanced Transportation prolonged driving was more likely to trigger the spatiotemporal tunnel efect, leading to the drivers' fatigue or even discomfort, weakening their perception of the driving speed, headway time distance, and tunnel width. Driving out of the tunnel onto a short-distance bridge not only afected the drivers owing to diferences in the external climate and environment but also frequently induced light and dark reactions. Te connection between the super-long tunnel and short bridge could easily cause the drivers to experience the illusion of driving and thus increase their driving load [41]. Terefore, to alleviate the driving load on the drivers on the test section, it was necessary to set clear direction guidance signs and marking lines on B1 bridge, to reduce the difculty associated with the direction identifcation, which reduced interweaving with trafc in other directions.

Fixation Behavior Analysis.
Te average fxation time for drivers familiar with the road section was short, ranging from one hundred and ten milliseconds to several seconds, and was infuenced by the difculty of the test and individual diferences. Fixations that are too short or too long may be uncomfortable or confusing, respectively [42]. Te fxation frequency and average duration for the middle tunnel during daytime driving were signifcantly lower than those for the entrance and exit, as shown in Figure 8  Journal of Advanced Transportation frequency and average duration for the tunnel entrance section were signifcantly lower than those for the middle and exit sections during nighttime driving, as shown in Figure 8 Te fxation duration for nighttime driving tended to be stable, the trafc fow at night was small, the surrounding trafc environment was familiar, and the fxation duration was relatively short. During the day, the fxation duration was the longest at the diverging section of B1 bridge, and the fxation frequency was the highest at the beginning and end of the test section, indicating that the drivers had to pay more attention at the beginning and end of the test section, to avoid accidents such as rear-end collisions. When driving into the entrance of tunnel T1 (slope, 6%), the drivers were prone to driving illusions, underestimating the true slope of the road, and not decelerating to the safe speed before entering the tunnel. At the end of the test section, the drivers experienced the space-time tunnel efect because sudden changes in the illumination environment at the tunnel entrance induced blind periods, causing the drivers to underestimate the actual driving speed, and thus fail to decelerate to the safe speed for driving on the main road. As a result, drivers were more likely to overspeed, creating a large speed diference between the tunnel trafc fow and the main road trafc fow. Along with the larger trafc fow on the main road and more complex lane changes, this contributed to the higher driving load.

Saccade Behavior Analysis.
Saccades are quick searches of the visual feld that serve to identify and pick up stimulusrelated information; the average duration of a saccade episode is 20-40 milliseconds [43]. Te more the saccades, the longer the search process. When a driver is fatigued, the perception of danger decreases, the number of saccades increases, and the ability to acquire stimulating  Journal of Advanced Transportation spatiotemporal information decreases. Tus, saccade characteristics refect the fatigue state of drivers to some extent. Te number of saccades and mean duration of saccades in the middle section of the daytime tunnel were signifcantly lower than those at the tunnel entrance and exit, as shown in Figure 9(a). At night, the section connecting B1 bridge to ramp and B2 bridge to T3 tunnel was associated with the highest number of saccades, and the search process was longer, as shown in Figure 9(b). Te number of saccades for nighttime driving fuctuated greatly and was signifcantly higher than those for daytime driving, indicating that the driving load associated with the bridge-tunnel section was higher and visual searching was longer during nighttime driving, which increased the drivers' fatigue.

Viewpoint Trajectory.
Te viewpoint trajectory can clearly represent the driver's fxation behavior characteristics of obtaining road and target information when driving [44]. In this study, the trajectory and distribution of drivers' fxation points in 3 s before and after diferent positions are analyzed.
In the daytime, the drivers' fxation points were mainly distributed in front of the speed dashboard and driving lane in the three-tunnel entrance and exit sections of the experimental section. In the middle of the tunnel, the drivers' fxation points are more discrete and mostly distributed in the distant area of the driving lane. In the bridge, drivers' fxation points are mainly distributed in the far front of the speed instrument panel and driving lane. At the diverging point of the ramp, the driver's fxation point is mainly distributed in the navigation area of the mobile phone and near the front area of the driving lane, as shown in Figure 10.
At night, the drivers' fxation points are mainly distributed in the front of the driving lane in the three-tunnel entrance sections of the experimental section. In the middle of the tunnel, the drivers' fxation points are more discrete and mostly distributed in the distant area of the driving lane. At the tunnel exit, drivers' fxation points are distributed   Saccade count (N)   Figure 11. Comparing the pilot fxation point distribution during the day and at night, the driver more often pays attention to the tunnel entrance section speed dashboard during the day than at night. Te reason is that the daytime running condition is good, the drivers are at a larger speed, which easily exceeds the speed limit at the entrance to a tunnel, and the driver should be more focused on the velocity by considering whether to brake, to ensure safe passage within the speed limit. At night, lighting conditions on ordinary road sections are limited, the driving speed is low, and drivers pass at low speeds in tunnel entrances and exits to meet the speed limit requirements. Terefore, attention to the speed dashboard is not high. In the middle section of the tunnel, the fxation points change in a consistent way. At night, the drivers pay more attention to the area near the front of the driving lane.

Conclusions
Te eye movement characteristics of the drivers, such as the pupil diameter, blink duration, fxation, saccade, and trajectories of fxation points, were analyzed using descriptive statistics and box graph methods. Te variation rules of the drivers' eye movement indices were obtained for the combined bridge-tunnel scenario. Te main conclusions are listed.
Te drivers frequently experienced light and dark reactions. When driving through long tunnels and short bridge convergence sections, the drivers could not reasonably control the driving speed. In addition, the drivers often experienced driving illusions. In light of the above, to improve the safety of nighttime on-ramp driving, it is recommended to install refective flms on both sides of the on-ramp, which is likely to improve the drivers' identifcation ability. It is suggested to set up corresponding road markings to indicate lane information at the long tunnels and short bridge convergence sections, to add lighting facilities and passive luminescence inducement measures, to set up recommended speed signs, and to limit vehicle speed reasonably.
It is necessary to set clear direction guidance signs and lines on B1 bridge, to reduce the difculty associated with the direction identifcation and the interweaving of trafc in the other directions when driving on on-ramp road sections. Tis study provides natural driving data for trafc departments to evaluate driving behavior and trafc safety control policies under the same scenario.
In this study, the combined bridge-tunnel scenario of driving on an expressway in a mountainous city was used as the experimental scene for obtaining relevant data, and the drivers' eye characteristic data were acquired at diferent times of the day and night. Te efects of the combined bridge-tunnel scenario on the drivers' pupil diameter change, blink duration, fxation, saccades, and vision trajectory were analyzed. However, this study had some limitations. Te study of the driving behavior in such scenarios allows to collect the drivers' physiological indicators, such as electroencephalogram and heart rate data, for improved driving in complex road scenarios.

Data Availability
Te data used to support the fndings of this study are available from the corresponding author upon request.

Conflicts of Interest
Te authors declare that they have no conficts of interest.