The Visual Scanning Behavior and Mental Workload of Drivers at Prairie Highway Intersections With Different Characteristics

Highway intersections are crash prone locations, and drivers’ improper attention allocation and sudden increase of mental workload are main contributing factors. To explore the visual scanning characteristics and mental workload of drivers at prairie highway intersections with different characteristics, an on-road driving test was taken at 6 intersections scattered on a typical prairie highway with 3 different shapes and 2 different priority rules, and drivers’ eye movement and ECG data were collected. The results show that there is significant difference in fixation allocation and transferring among intersections with different shapes and priority rules. The X-shaped intersection without the right of way shows the longest average fixation duration and the most gaze shifts between the drivers’ lane and the intersection road, whereas the three-way intersection with the right of way shows the shortest fixation duration and the gaze transfer path is relatively simple and the transfers are less. The mental workload implies almost the same conclusion, with the X-shaped intersection without right of way shows the highest mental workload which is manifested by largest heart rate growth rate and the lowest RMSSD, whereas the three-way with priority of way shows the opposite trend. This study revealed that both intersection types and priority rule made differences in drivers’ scanning behavior and mental workload at prairie highway intersections, which can support the development of appropriate countermeasures regarding the applications in advanced driver assistance systems and the design of intersection constructions to warn drivers about potential critical events at different types of intersections.


I. INTRODUCTION
Road intersections are arguably crash prone locations due to the higher situation awareness requirement caused by the complex traffic environment [1], [2], [3] [4], [5], [6]. Up to 60% of severe injuries in traffic accidents occurred at intersections in European countries [5], [7]. In the United States, 47% of car crashes occurred at intersections or The associate editor coordinating the review of this manuscript and approving it for publication was Miaohui Wang. intersection related junctions in 2015, and these intersection crashes contributed to over 50% of crash injuries (NHTSA, 2015). Given the complexity of geometric features and traffic from multiple directions, intersection negotiations often involve significant speed differences and higher rates of non-stopping traffic on the major roads [8], [9], [10], successful intersection maneuvers require the ability to perceive all useful information resource and make decisions in such a dynamic driving environment. So, the visual scanning behavior and the change of drivers' mental workload at VOLUME 10, 2022 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ intersections are of great importance in understanding the driving behavior and the traffic accident mechanism at intersections [7], [11]. It is found that drivers with scanning problems had 4.2 times more collisions and 15.6 more intersection accidents than those without this problem [12]. One of the main contributing factors with these intersection accidents is improper attention allocation [13]. So, drivers' visual scanning behavior is critical for the safe driving at intersections [2], and a lot of research was done on this aspect [5], [14], [15], [16], [17], [18], [19], [20]. Wickens et al proposed the SEEV model of information acquisition in visual scanning to identify the role of event salience, effort, expectancy, and value in influencing where and when people look at different channels to sample information in dynamic environments, such as aviation and driving, Fig.1 gives the four components of the SEEV model [13], [20]. Bremond's research on the gaze behavior when approaching an intersection in a driving simulator suggested that visual attention to intersecting roads varied with the priority rule [7]. Li et al collected the eye movement data of the driver's intersection under the natural driving state, and compared the difference in the visual characteristics of drivers at the signal-controlled intersection and the uncontrolled intersection, and they found that intersection types did make differences on drivers visual scanning behavior [2]. Costa made comparison of driver's yielding behavior with different T-junction priority rule and found that a significant speed reduction and an increase of driver's visual inspection to the intersection area in the priority-to-straightarm condition in comparison to the priority-to-intersectingarm condition [21]. The difference of visual behaviors with different maneuvers at intersections was compared in some papers. For example, Shan B found that drivers checked for traffic significantly more often before they went straight across than for proceeding with a left turn [19]. Romoser's study also showed that the average amount of time spent in different regions is different when drivers made right, left turn, and went straight [22]. Yamani et al. found that drivers paid little attention to potential threats on the roadside when passing through uncontrolled intersections [18]. The research of Li et al found that driver's viewpoint path was linearly distributed, and the gaze density and gaze intensity of conservative drivers were significantly higher than those of aggressive drivers when turning right, and aggressive drivers ignore the right area when turning right [23].
Some researchers intensively analyzed driver's visual scanning behavior at intersections with the effects of age and driving experience [24]. Bao and Boyle's study on the visual characteristics of drivers of different ages at intersections found that older drivers paid more attention to the area in front of the vehicle at the intersection, and less attention to the left and right areas [25]. The research of Guo Fengxiang on the visual characteristics of older drivers in road intersections based on driving simulation showed that the older drivers' attention to the gaze area in the non-driving direction was relatively poor, and the flexibility of the gaze transfer mode was relatively poorer than that of the young and middleaged drivers [26]. Yonekawa studied the driving behavior of elderly drivers passing through stop sign intersection using driving simulator and found that the left/right safety checks of the elderly drivers were not sufficient [27].
There were also some studies on visual behaviors of drivers at intersection in some special circumstances. For example, Liu et al. studied the relationship between drivers' speed and visual characteristics at intersections with consideration of risk perception [28]. Xiong examined drivers' cell phone use behavior at intersections by using naturalistic driving data [29].
The complexity of the driving environment can affect the information processing and attention of drivers [30], [31]. Heger found that the mental capacity of drivers reduced with the increase of their information input in the driving environment, resulting in the feelings of pressure and stress [32]. The driver's mental workload refers to the driver's sense of tension and pressure when performing corresponding driving tasks or non-driving-related tasks under the influence of the external traffic environment and internal cognitive state. It can lead to errors in judgment, decision-making and operation of the driver, resulting in traffic accidents [33], [34]. Many research on driver safety and the driving mental workload confirm that accident risks and drivers' mental workload are strongly associated [35], [36]. If the mental workload exceeds driver' information processing capacities, it may result in wrong judgement or even improper maneuver, which may cause an accident. [37], [38]. The measurement methods of the driver's mental workload include subjective evaluation method, driving task performance method, physiological parameter evaluation method and so on [31], [33], [39], [40], [41]. Studies have found that when a person's mental load changes, the related physiological indicators will also change, so the physiological parameter evaluation method can provide real-time, objective and highly sensitive evaluation of mental load [31], [42], [43]. Partin suggested that heart rate respiration rate sensers were valuable in observing a driver's awareness [40]. This was verified by Healey and Picard in detecting drivers' condition during real driving [44]. Miller found that HRV was an effective measure of mental workload and it was also used by many studies in real-world driving [34], [45], [46]. Brookhuis did research on the difference in the psychological pressure of drivers when driving on a congested circle of urban roads and unobstructed roads sing electrocardiogram, and found that in an alert state, drivers are more nervous when driving on the circle than on the highway [47]. In Veltman's study, the sensitivity of physiological measures to mental workload was investigated in a flight simulator, and all measures, which include heart period, continuous blood pressure, respiration, and eye blinks, showed differences between rest and flight, and between the pursuit and the tunnel task [48]. Guo Fengxiang et al. studied the physiological characteristics of drivers in the intersection and found that the drivers' growth rate of heart rate increased and the RMSSD of HRV indicator decreased, and the age of drivers showed significant effect [26]. The result of Hao's study showed that the average heart rate, and LF/HF ratio increased significantly with the increase of traffic density [49]. Liu's study identified the heart rate and the pupil diameter were the most representative physiological parameters of drivers' mental workload when they were passing through highway intersections. The larger the increase of heart rate, the heavier mental workload and the smaller the pupil diameter of the drivers [50].
It can be seen from above, most of the research on the visual perception and mental workload of drivers mainly focused on typical rural and urban intersections, and most of which were carried out based on virtual driving scene platforms. The landscape of typical prairie highway driving environment is significantly different from other road driving environments. It is mainly composed of flat grassland with few humanistic landscapes, and the landscape color along the road is monotonous. Furthermore, the terrain and morphology changes on the whole line are small, which is represented by the large radius of the vertical curve (the longitudinal slope is mostly concentrated below 2%) and the monotonous horizontal alignment dominated by long straight lines connected with large radius curves (the length of the straight line is about 80%, and the radius of the flat curve is more than 4000m for more than 60%) [51]. So, drivers are in a low mental workload driving state most of the time during driving [51]. But when an intersection appears ahead, the drivers' information perception and cognitive demand will increase instantaneously, which may result in disorder of drivers' attention allocation and sudden increase in mental workload. According to the investigations on the prairie highways, nearly 1/3 of the accidents occurred at the intersections [52]. Previous studies on prairie highways have showed that drivers' visual perception characteristics and mental load are different at various driving conditions such as intersections, curves and overtaking maneuvers compared with the straight sections, but they only focused on the typical Crossshaped intersections and the indicator of the mental workload was only eye movement data [53], [54]. Both the types and traffic control rule factors of the intersection were not considered. So, it is of great necessity to do research on the visual allocation and mental workload of drivers at intersections of the prairie highways with different characteristics.
According to the survey result of several typical prairie highways, the common intersection shapes include Crossshaped, Three-way, and X-shaped with the priority right of way or yield. So, a typical prairie highway section with all these shaped intersections was chosen and a real vehicle test was conducted. The eye movements and ECG data of drivers at these typical intersections were collected to analyze the attention allocation and mental workload characteristics of drivers at all these kinds of intersections.
Our main goal is to understand how the intersection characteristics, the types of intersection and the traffic rules at intersection impact the drivers' allocation attention and mental workload during the intersection passing on prairie highway. It is hypothesized that the drivers have longer average fixation time and reallocate their main attention frequently when they are confronted with the more complex X-shaped intersection and their mental workload will increase dramatically, whereas the Three-way intersection shows the lowest fixation time and simplest attention allocation and the least change of mental workload. We also predict that the priority rule at the intersection will affect the driver's vision characteristics and mental workload, and the intersections without the right of way draw more attention and result in higher mental workload.
The remainder of this paper is organized as follows: section 2 describes the methodology of the test experiment. Section 3 gives the results for attention allocation characteristics and mental workload of drivers at different types of intersections. Section 4 is a brief discussion of relevant issues. Section 5 draws some conclusions and presents further research.

A. PARTICIPANTS
The sample size was calculated based on the index of expected variance, target confidence and margin of error [55], and the calculation formula of which is shown in Equation (1).
where N represents the sample size, Z represents the standard normal distribution statistics, E represents the standard deviation, and E is the maximum allowable error. In this paper, the significance level is taken as 10%, then Z=1.25. Due to the limitation of the number of subjects in the real driving test, the given value of σ is taken as 0.4(the value range of it is between 0.25 and 0.5), and E takes 10%.
According to the Equation 1, the minimum sample size is 25, so 28 drivers were recruited as subjects and informed about the experiment's general purposes. All subjects declared that they have statutory driving licenses with normal eyesight, and the average age of which is 39.5 and with more than three years (mean= 6.3, SD=5.2) of driving experience. All the subjects were required to be in good physical and mental state before the test and they all furnished explicit consent about data recording of their eye movements and ECG.

B. EXPERIMENTAL EQUIPMENT
The main equipment of this test is shown in Fig.2. Since 80% of vehicles on the prairie highway are passenger cars, a black Passat sedan was used as the test vehicle, as is shown in Fig.2(a). An I view X HED helmet-mounted eye tracker was used to collect eye movement data (see Fig.2(b)) and the Begaze software matched with the eye tracker was used to extract and analyse the data. The sampling frequency of the eye tracker was set at 200 Hz/s to ensure the reliability and the integrity of the data collection while drivers passing through the intersection. The driver's ECG index was collected by an MP150 Multichannel physiological instrument (see Fig.2(c)). Apart from these equipment, some other auxiliary equipment was used to ensure the continuity of the test and the complete record of the data which include laptop computers, largecapacity batteries, 12 V DC batteries, etc.

C. THE ROAD FOR EXPERIMENT
The test road is a section of a typical prairie secondary highway with a total length of 150 km, and the start and end point of which are Sai Han Ta La and Man Du La Tu respectively. The basic road and traffic information of the test road is shown in Table 1. According to the survey on the typical prairie highways, most of the intersections on the way are cross-shaped intersections, three-way intersections, and X-shaped intersections, among which three-way intersections account for 35%, and the cross-shaped and X-shaped intersections account for 35% and 30% respectively. Since the intersections along the way are all unsignalized and the priority rule before the intersection include priority and yield, the intersections can be classified as intersections having priority and intersections having to give the right of way. So, the intersections can be further divided into 6 types according to both shapes and priority rule. The classification of intersections is shown in Table 2, and the schematic diagram of various intersections is shown in Fig.3.

D. EXPERIMENTAL PROCEDURE
According to the classification of intersections above, this study uses a 3 × 2 within-subjects factorial design with the intersection shape and priority rule as independent variables, and drivers eye movement and mental workload as dependent variables.
To eliminate the influence of external environmental conditions on the equipment and the driver during the test, the test was carried out in September when the road landscape colour and climate conditions are consistent. The test is divided into two stages. The first stage is a 5 min static test and a 5 km driving adaptation training to ensure that the subjects can adapt to the driving environment and equipment to get into a normal driving state. The second stage is the formal test. In the formal test, the subjects are arranged to drive straight along the test road section, and to cross the intersections avoiding collisions and following the traffic rules. To minimize the effect that may occur due to the type of intersection that was encountered first, half of the participants drove the route in a clockwise direction, while the other half drove in a counter-clockwise direction. All the electronic devices in the car that are not related to the test should be turned off and chat is not allowed. The total duration of the test is about 2 h.

E. INDICATOR SELECTION AND DATA EXTRACTION
Since divers' fixation behaviours are generally considered to be effective to reflect the collection of road traffic information and identify the driver's intention., and the blinking behaviour and ECG of drivers are often used to evaluate driving fatigue and driving load [31], [39], [40], [41], [42], the attention allocation characteristics is measured by fixation duration, the number of fixations and the transferring of it from one interest of area to another, and the mental workload is determined using the blink duration, the growth rate of heart rate and the RMSSD, a time domain indicator of HRV in this study. Table 3 presents the description of all these dependent variables.
To extract the data in calculating the eye movement indicators above, the Begaze software of the I view X HED eye tracker was used to replay the entire eye movement video of each subject, and the eye movement data of the subjects driving through the six intersections were extracted. To analyse the visual behaviour performance of the driver passing through the intersection comprehensively, 10 s is taken as the time length of the eye movement data interception of the intersection in this study, that is, the eye movement data was extracted from 10 s before every kind of intersections. The Acknowledge software is used to intercept the driver's ECG data in 6 types of intersections. Considering the lag of the driver's emergency response at the intersections, a total of 15 s, from 5 s before the intersection to 10 s after it, is used as the interception time of the ECG data of drivers at intersections.

A. THE FIXATION CHARACTERISTICS OF DRIVERS AT INTERSECTIONS 1) DIVISION OF AREAS OF INTEREST (AOI)
To study the driver's attention allocation characteristics and gaze transfer patterns when they are driving at different types of intersections, it is necessary to divide the driver's gaze area according to their visual interests. There are many methods for dividing the region of interest, such as mechanical division method, frame-by-frame statistical method, dynamic clustering method, etc. [19]. Based on the driver's visual characteristics at various prairie highway intersections, the driver's areas of interest (AOI) are divided into four in this study, which include the drivers' lane, the intersection road, the In-vehicle area, and other areas in-side and outside of the vehicle. The division of areas of interest is depicted in Fig.4 and the abbreviations of them are in table 4.   The percentage of fixation duration is defined as the duration of fixation points in one area divided by the duration of fixation points in all areas, as shown in Equation (2).
where i represent the number of gaze points, j represents the region of interest, α j refers to the ratio of the driver's gaze duration to the region of interest j, and t ij represents the gaze duration of the i-th gaze point in the region of interest j. The percentage of fixation points is defined as the number of fixation points in one area divided by the number of fixation points in all areas, as shown in Equation (3).
where β j refers to the ratio of the number of fixation points to the area of interest j, and N j represents the number of fixation points of the driver in area of interest j. Fig.5 and Fig.6 show the comparison of the fixation duration, the numbers of fixations and the ratio distribution of the fixation duration and the fixation point to different areas of interest at different intersections.
It can be seen from Fig.5 that the T&P intersection shows the smallest average fixation durations, whereas the X&Y shows the largest. The two-way repeated ANOVA shows the difference among different intersections is significant. Compared with the intersections with the right of way, drivers tend to allocate more attention to the intersection road without the right of way, and the percentage of the fixation duration to the intersection road is the largest at the X-shaped intersection.

b: NUMBER OF FIXATIONS AT DIFFERENT INTERSECTIONS
It can be seen from Fig.6 that the T&P intersection shows the most fixation times, whereas the X&Y shows the least, which  is contrary to the feature of the fixation duration. The twoway repeated ANOVA shows the difference among different intersections is significant. Compared with the intersections with the right of way, drivers tend to pay more attention to the intersection road without the right of way, with more fixations to the intersection road, and the percentage of the fixation duration to the intersection road is the largest in the X-shaped intersection.

3) THE FIXATION TRANSFER CHARACTERISTICS AT DIFFERENT INTERSECTIONS
A probability vector defined by Wang [56] is used in this study to gather gaze transfer probabilities between four AOIs (DL IR IV OA) as is shown in Fig.7. The proportion matrix in this study has 16 elements, referring to the 4 × 4 different consecutive pairs of transitions displayed in Table 5.
Each element is the proportion of saccades moving from location i to j. If i=j, it is the proportion of saccades in the location. We let n ij be the number of pairs (i, j); and 4 i=1 4 j=1 n ij be the number of total saccades. Then, P ij , the proportion of saccades from location of i to location j, was defined  in Equation (4).
Two essential requirements of the proportion matrix in this study are listed in Equation (5) and (6) as following: The gaze transfer probabilities of drivers at intersections with different shapes and priority rules are statistically calculated according to the Equation 4 and the heat maps of gaze transition pattern with proportions is plotted and shown in Fig.8.
The attention allocation of drivers at intersections with the right of way are shown in Fig.8(a), Fig.8(b) and Fig.8(c). The three intersections with different shapes show different features. The C-shaped and X-shaped intersections show larger radian at the top of the three-dimensional surface, whereas the Three-way intersection's radian is very small. That is because the gaze transferring between the DL and IR at the C-shaped and X-shaped intersections is much more than the Three-way intersection, and the total transferring proportion of fixation between the DL and IR are 87.86% and 91.36% respectively compared with the 83.41% at Three-way intersection. Apart from this, we found that the gaze transferring inside the DL is highest at Three-way intersection with the proportion of 33.83%, and it is 29.35% and 28.59% at the C-shaped and X-shaped intersections respectively. Since the DL and IR AOIs are the main areas for key information acquisition at intersections, the fixations and gaze transferring between other AOIs except from the DL and IR is rare especially at X-shaped intersection the ratio of which is less than 9%.
The attention allocation of drivers at intersections without the right of way is like the features of intersections above with larger radian at the top of the surface. As is shown in Fig.8(d), Fig.8(e) and Fig.8(f), the X-shaped intersections shows obviously increasing attention to the intersection road compared with C-shaped and Three-way intersections. The probability of the gaze shifting from the DL to the IR and from the IR to the DL are 24.42% and 23.53% respectively at X-shaped intersection without the right of way, and fixation transferring inside the IR AOI is 18.21%. Whereas the percentage of the fixation transferring inside the DL declined from 28.59% to 27.64%. The change of C-shaped intersections is slightly small, and the total transferring percentage to the critical areas (DL and IR) is 92.5% compared with 93.8% at X-shaped intersection. This ratio is the least at three-way intersection (84.75%). The proportion of gaze transferring inside the DL is 30.49% at Three-way intersection, and it is still the largest compared with C-shaped and X-shaped intersections.
When the traffic rule factor is considered at intersections with the same shape, taking the C-shaped intersection as an example, we can find that drivers' attention allocation is more scattered and the gaze point transfer path is relatively fixed when they do not have the right of way. Since drivers need to acquire information from the intersection road to make decision for driving speed, their gaze transferring between the driving lane and intersection road is larger at the intersection without the right of way. This is manifested by the larger radian at the top of the three-dimensional surface. The Threeway and X-shaped intersections show the same features, but the difference of Three-way intersection with different priority rule is more obvious, because the information acquisition demand is not as much as C-shaped and X-shaped intersections. Also, the percentage of drivers' gaze shifting inside the DL AOI is lower at intersections where drivers must yield the way, because of the more visual information need to the intersection road. In general, drivers must pay more attention to the most important information areas due to the large demand perceive more information for decision-making and corresponding operational behaviour.

B. THE MENTAL WORKLOAD OF DRIVERS AT INTERSECTIONS
Many research results show that the blink behavior of drivers' eye movement and the heart rate growth rate and heart rate variability (heart rate variability, HRV) of the ECG signal are effective indicators in the driver's mental workload [31], [33], [41], [57], [58], [59].So, the blink duration of the eye movement, the heart rate growth rate (HRGR) and the RMSSD in the HRV time domain index are selected to analyze the change of drivers' mental workload at intersections VOLUME 10, 2022 of prairie highways with different shapes and priority of way.

1) ANALYSIS OF BLINK DURATION
The blink duration refers to the time taken by the subjects' eyes from fully open to closed and to fully open again, which can reflect the subjects' cognitive load [57]. The shorter the blink duration, the greater the cognitive load of the subjects [58], [59].
To analyze the distribution of blink duration, we divided blink events in 10 ms duration classes and Gaussian-curve was obtained. As is can be seen from Fig.9, in yield conditions, the distribution of blink duration shifts towards shorter blinks (the red straight lines compared with the black straight lines in the Fig.9) no matter what the shape of the intersection is, and the shift of Three-way intersection is the largest. When the shapes of the intersection are considered, it can be found that more short blinks did occur at the X-shaped intersections whether the drivers enjoy the right of way or not, and the Three-way intersection shows larger blink duration.

2) ANALYSIS OF HEART RATE GROWTH RATE (HRGR)
The heart rate growth rate is defined as the difference of heart rate between the real-time during driving and in a calm state divided by the heart rate in a calm state, as is shown in Equation (7).
where H JS refers to the real-time heart rate value during driving, and H JC is the heart rate in a calm state. Two-way repeated measures ANOVA is used to analyse the difference of the heart rate growth rate at intersections with different shapes and priority rule. Table 6 shows the descriptive statistics for HRGR, and a multivariate analysis is used to detect repeated-measures effects. The results showed that both the intersection shape and priority rule reach statistical significance (F(2,26)=14.471, p<0.05, F(1,27)=22.782, p<0.05), whereas the shape×priority rule is not significant. The comparison chart of HRGR at different intersections is shown in Fig.10. Bonferroni tests are further applied to compare the sources of differences. The results show that there are significant differences among intersections with all three different shapes and drivers' heart rate growth rate is the highest at the X-shaped intersection followed by C-shaped intersection. The Three-way intersection shows the smallest growth rate.
The priority rule of the intersection has influence on the drivers HRGR, and the intersections without the right of way shows higher HRGR no matter what the shapes of intersections are.

3) COMPARATIVE ANALYSIS OF RMSSD
RMSSD refers to the root mean square of the difference between two adjacent R-R intervals, which is used to estimate the components of short-range HRV, and the lower the RMSSD value, the higher the mental workload. It is calculated as Equation (8).
where t RR,i and t RR,i+1 are the lengths of two adjacent sinus cardiac cycles. Table 7 shows the descriptive statistics of RMSSD at intersections with different shapes and priority rules. The multivariate analysis is used to detect repeated-measures effects and the results showed that the intersection shape and priority rule both reach conventional levels of statistical significance (F(2,26)=125.874, p<0.05, F(1,27)=20.762, p<0.05), whereas the shape×priority rule is not significant. The comparison of RMSSD in different intersections is shown in Fig.11. Bonferroni tests are further used to compare the sources of differences. The results show that there are significant differences among intersections with three different VOLUME 10, 2022   shaped intersections, and drivers' RMSSD is the smallest in the X-shaped intersection whereas the Three-way intersection the largest. As is seen from the Fig.11, the RMSSD of drivers is smaller at intersections where the drivers do not have the right of way.

4) MENTAL WORKLOAD COMPARISON BASED ON ANDREWS VISUAL HARMONIC CURVE
To visually illustrate the differences of drivers' mental workload at different types of intersection under the comprehensive influence of the three indicators above on a twodimensional plane, the Andrews Curve is used to reduce the dimension and visualize it [60], [61]. The principle of the Andrews harmonic curve is as follows: Multivariate data can be represented in the form of a vector. For example, a p-dimensional vector can be represented as a p×n multivariate data set with p observation indicators, and each indicator has n sample observations. The p indicators of each sample constitute a point in the p-dimensional space, which can be represented by a curve in the two-dimensional space. Andrew's curve defines a finite Fourier series as Equation (9).
where t ∈ [0, 1] and mod(k,2) is an operation returning the remainder after division of k by 2 to ensure that each data can be viewed as a line between 0 and 1. If the Andrews harmonic curve of one intersection is obviously different from the curve of another intersection, it indicates that this visualization tool is effective to describe driver's mental workload at intersections by jointly considering the information from multiple aspects both from eye movement and ECG of drivers. MATLAB is used to draw the Andrews harmonic curves of drivers' mental workload under the combined action of the three indicators including blink duration, HRGR and RMSSD at various intersections, as shown in Fig.12.
To reflect the differences of different intersections simply and clearly, only the mean value (the solid line with label in the Fig.11), the 25% and 75% quantiles (dashed lines on both sides of the solid line) are displayed. It can be seen from the Fig.12 that drivers' mental workload can be clustered into four patterns. The T&P and X&Y belong to two different patterns. The C&P and T&Y can be grouped into one pattern since the feature of the two curves are nearly the same. Similarly, the C&Y and X&P belong to one pattern. So, the Andrews visual harmonic curves show the similar feature to the previous mental workload analysis results and it can reflect the difference of mental load at different intersections intuitively.

IV. DISCUSSION
The safe driving at highway intersection requires for reasonable attention allocation and moderate mental workload to visually detect information and make correct decisions [16], [62]. Selective attention directs gaze towards objects of interests and concerns in the environment, and the mental workload of drivers increase instantaneously for these extra efforts [63]. Therefore, differences in visual selection and mental workload may reflect drivers' different strategies and the efforts they need to pay at different types of intersections with different control rules.
The aim of this study was to examine the difference of visual characteristics and mental workload of drivers at different shapes of intersection on prairie highways with various control ways. The visual perception differences were assessed using proportion of fixation duration and fixation number of those drivers at different AOIs and the transferring of gaze between them at these intersections in this study. The results confirm the hypothesis that differences do exist among these intersections with different shapes and priority rules in terms of where they look and for how long, and the gaze transferring between different AOIs are also different.
In the 'Priority' condition, although the drivers have the right of way, they still need to allocate some attention and spend some time to the intersection road to find out whether there are cars coming to the intersection and whether they stopped or not. Since drivers' vision is limited at X-shaped intersections, their attention allocation to the intersection road is significantly more than the cross-shaped and three-way conditions.
In the 'Yield' condition, the driver needs more and real information from the intersecting road AOI to estimate the gap between consecutive vehicles running on the intersecting road [64]. So, the percentage of fixation time and the number of the fixations to the intersection road are larger compared with the 'priority' condition. Mradiano Costa et al examined two different priority arrangements applied to T-junctions (priority-to-straight-arm condition and priorityto-intersecting-arm condition), and checked drivers' effects on visual inspection of the intersection area. The eye movement analysis showed that total fixation time towards the intersection critical area and horizontal eye movements were significantly higher in the priority-to-straight-arm condition. The conclusion is similar to the study on the Three-way intersection with different priority rules [21]. Sophie Lemonnier et al studied on the gaze behavior when approaching an intersection, and their results suggest that the driver's fixations were mainly distributed between the driver's lane and intersecting road AOI (96.69%) in all conditions. More attention is allocated to the intersecting road AOI in the 'Yield' condition than under the 'Priority' condition, and the smallest dwell time was found in the 'Stop' condition [7]. So, the conclusion is partially our conclusion. Since most of the control ways at intersections on the prairie highways are 'priority' and 'yield', whereas the 'stop' sign is rare, the 'Stop' condition is not considered in our study.
Apart from the priority rule, drivers' attention allocation varies with the intersection shapes. The average fixation duration and the number of fixations is larger at X-shaped intersections, and the percentage of them to the intersecting road area are larger than C-shaped and Three-way intersections no matter drivers enjoy priority of way or not. Also, drivers' gaze transferring path shows more disorder compared with other two shapes which is manifested by the larger radian of visual transfer surface. Julia Werneke examined the influence of two environmental factors (traffic density and information volume) in a T-intersection situation on the drivers' attention allocation in a driving simulator study, and the result showed that the participants paid significantly more attention to the left-hand side at intersections with high traffic density [17]. In our study, the AOI in front area was not further divided into left-hand and right-hand, and the traffic density and information factors are not considered either. Also, in some studies, the age difference of the eye movement is considered at the intersection [16], [18]. So, it would be interesting to consider these factors in future research.
The SEEV model proposed by Wickens et al explains visual search and attention allocation in general [13]. Our research result fits with Wickens' SEEV model that all visual VOLUME 10, 2022 attention resources are optimally allocated to the task relevant AOIs (driver's lane and intersection road). The Effort associated to transitions from Driver's Lane to Intersecting Road (and back) is smaller in the 'Priority' conditions, compared to the 'Yield' conditions. Also, it is smaller in the less complex Three-way and C-shaped intersections, and further study can be done to verify the fitness of SEEV model on the prairie highways.
The mental workload was measured by drivers' eye blink duration, heart rate growth rate and RMSSD of ECG in this study. Differences were found among various types of intersection with different kind of control way, which confirm the second hypothesis. The result shows that the distribution of blink duration shifts towards more shorter blinks and the HRGR is the largest whereas the RMSSD lowest at X-shaped intersection without right of way, which means the X-shaped intersection where drivers must yield the way may bring more mental workload to drivers. Some studies have shown that the effect of the types of traffic conflict on the physiological characteristics of drivers are different [44], [45], [46], [47]. PHYU investigated the effect of roadway conditions on driving stress in Myanmar by using heart rate variability (HRV) and found that highly crowded places and those requiring attention are the most stressful segments along the roadway for drivers [44]. Liu's study identified the heart rate and the pupil diameter are the most representative physiological parameters of drivers' mental workload when they are passing through highway intersections, The larger the increment of the heart rate the heavier mental workload beard by the drivers and the smaller the pupil diameter of the drivers [50]. All these results are consistent with the conclusion of this paper.

V. CONCLUSION
Our research focused on the comparison of drivers' attention allocation and mental workload at prairie highway intersection with different shapes and priority rules. The results revealed that both intersection types and priority rule made differences in drivers' scanning behavior and mental workload. The X-shaped intersection without right of way showed the most obvious disorder and the largest mental workload whereas the Three-way intersection with the right of way shows the least complexity and the smallest mental workload. The most important practical implication of the current results is that it provides the salient visual and mental workload features of drivers at various intersections at prairie highways. When coupled with further validation, designers may be able to use this result to predict the allocation of visual attention and mental workload at different highway intersections and thereby predict vulnerability to miss roadway hazards. This is especially important as nearly 1/3 of accidents on the prairie highway happens at these intersections.
A limitation of the present study is that it only emphasizes on the visual and cognitive aspects of drivers, further research is needed on the driving behavior of drivers to illustrate the influence of visual and cognitive state on the driving maneuver of divers. Furthermore, as is described in the literature review, some studies have showed that age (young, middle aged, older) and turning maneuvers (left, straight, right) would influence drivers' visual scanning and mental workload simultaneously [25], [26], [63], we only focused on the visual scanning and mental workload at intersections with different shape and priority rule, and no comparisons among the age and turning maneuvers difference were examined. Future efforts would be concentrated on these aspects.