Abstract
Aim of the study is to investigate, by means of a driving simulator experiment, drivers’ performance in terms of lateral position, speed, deceleration, steering angle, and breaking times on a divided two-lane rural highway in relation to drivers’ visual attention (VD). In the experiment, the virtual scene of twenty different geometric alignment sections without traffic and the VD testing were designed. Twenty-three experienced drivers with the calibration of attention capacity participated in a 30 km drive in an interactive fixed-base simulator. Each participant was required to drive with the controlled speed of 60 km/h along the central lane as repeating random number and was evaluated on VD and driving performances. Three different data analysis techniques were used: (a) statistical tests and hypothesis test of curvature change rate (CCR) of the geometric alignments, visual attention demands, and driving performance data, (b) correlation analysis of VD, CCRs, and driving behaviors, and (c) regression analysis of the VD and CCRs. Results have showed that the driving performance can be effectively influenced by the highway alignment and a prediction model built in this study can evaluate the drivers’ visual attention demands before the highway constructed. The interactions among VD, driving behavior, and CCRs were also found.
1. Introduction
Driving requires continuous and high level concentration of attention. Attention is the taking possession by the mind in clear and vivid form of one out of what seem several simultaneously possible objects or trains of thoughts. It implies withdrawal from some things in order to deal effectively with others [1]. Inattention during driving and driver distraction account for a substantial number of traffic accidents [2, 3]. One study showed that the accidents related to drivers’ inattention as a contributing factor have been in the range of 10% of all accidents [4].
Driving is a certain risk behavior that is highly vision-dependent [5]. Safe automobile driving requires drivers to process large amounts of dynamic information under time pressure. When someone is driving a car on the highway, his operating behaviors are mainly determined by his visual judgment of the highway's alignment. The visual shape of the travel space directly affects the driver's driving behavior and speed. Most researches on driver's visual attention have analyzed drivers’ visual behavior and driving behavior [6, 7]. Drivers’ glancing and driving behaviors reflected the visual attention demands when drivers carry out nondriving secondary tasks, and significant research has either focused on one or used both as indicators of driver attention demands. Wierwille proposed a simple model which describes the change of gaze observable while driving with visual secondary tasks [8]. A model by Salvucciand on attention processes in multitasking was used to model attention in driving with secondary tasks. They simulated gazing and driving stabilization behavior for several secondary tasks. This model takes into account only the stabilization level of the driving task (steering and acceleration/braking) with a secondary task [9]. Driving behaviors, including both the longitudinal control measures (e.g., speed variance and mean speed) and lateral tracking measures (e.g., lane position, lateral speed, and steering wheel angle) are valid bases for evaluation and have been widely used in studying drivers’ attention resources. A number of simulator studies have examined the driver's dynamic visual characteristics [10, 11] and mostly based on eye-tracking test technology or road driving simulation experiment [12, 13].
Some curved sections of urban motorway have comparatively high frequency of accident occurrence. Hong et al. attempted to link and grasp three factors, namely, eye movement, roadway alignment, and driver speed and steering control and proposed a model that evaluates the risk of curved section in relation to the driver's eye movement and driving behavior [14]. Design consistency refers to the conformance of a highway's geometry with driver expectancy. A consistent design allows drivers to perform the task of driving safely, allowing attention or capacity to be dedicated to obstacle avoidance and navigation. Tsimhoni and Green found visual demand increased significantly with the reciprocal of curve radius [15]. Fitzpatrick et al. found speed variance was inappropriate as a design consistency measure for horizontal curvature, and visual demand was a measure of the information processing demands imposed by roadway geometry on a driver [16]. Wooldridge et al. also revealed several relationships between roadway geometry and visual demand [17]. Curve radius and its reciprocal were found to be significantly related to visual demand. Based on the model of driver visual demand on simple horizontal curves, Easa and Ganguly modeled driver visual demand on complex horizontal alignments that included simple, compound, and reverse curves [18]. On the following Easa and He developed models for evaluating visual demand on 3D highway alignments to carry out a design consistency evaluation [19].
To investigate the correlation among highway alignments, visual attention demands, and driving performance, we had posed the following experiment which was designed to further investigate the impact of roadway alignment on driver performance and driver’ visual attention demand. The primary questions of interest were as follows. (1) What are the changes in drivers’ visual attention demands on rural highways with different curvature change rate (CCR)? (2) How are these changes reflected in driving behaviors? (3) What is the correlation between rural highway design consistency and drivers’ visual attention demands?
2. Methods
2.1. Participants
Twenty-three participants (7 females and 16 males) aged from 25 to 58 years old (M = 36.2, S.D. = 9.7) took part in this study. All held valid license and were experienced drivers (driving experience from 5 to 21 years, M = 7.2, S.D. = 5.3) who drove at least 8,000 km annually. The participants were selected from a test driver panel of our laboratory. All drivers had completed at least 20 minutes of practice in the driving simulator and had mostly participated in several other driving simulator studies before. All had normal or corrected-to-normal visual acuity and did not take any kind of medicine.
2.2. Apparatus
The experiment platform built by the simulation laboratory of Faculty of Transportation in Kunming University of Science and Technology includes a driving simulator and a visual attention demand system.
2.2.1. Driving Simulator
The experiment was conducted with a full-size interactive driving simulator (KMRTDS) with fixed base (Figure 1). The driving simulator setup consists of a fully equipped XiaLI sedan, which featured all normal displays and controls (steering, brakes, and accelerator), placed in front of an arc screen subtending 150° horizontally. The virtual environment was projected on the screen at a resolution of 1024 × 768 pixels from three projectors at a frequency of 50 Hz. The simulator rendered dynamic images of driving scenarios based on the inputs of drivers. The participants in the car operated the controls, moving through the virtual world according to his or her inputs to the car. This system provides realistic road, fog, and dynamic traffic flows with appropriate direction, speed, and intensity. To test absolute validity, comparison between real-world and driving simulator speed data was performed in a previous study. Before the experiment we ensured the real-time of the system and the fidelity of the experimental scene.
The KMRTDS driving simulator.
2.2.2. Visual Attention Demand Test
Visual attention demands test is based on a special performance of visual occlusion. The driver must attend a secondary visual task developed by software on a computer under “safe driving” process. The secondary visual task will ask the driver to transfer the vision from the driving context to the secondary visual task, which forms the visual occlusion for effective driving information of the highway geometric alignments, traffic flow, and so on. (Figure 2 showed the secondary visual task on the laptop computer laying right side of the driver's line of sight.) The “safe driving” is defined as the test vehicle should not collide with other vehicles or exceed the driving lane edges at any time.
The visual attention demand test while driving and the software interface.
The secondary visual task is designed by random number repetition. Subjects drove the test track whilst repeating random numbers generated separately in the 12 text-boxes on the software interface. The percentage of numbers repeated whilst driving the test track was compared with the percentage repeated whilst the vehicle was stationary. To calibrating the subjects’ attention capacity, the subjects should repeat correctly when the vehicle is stationary. Let the percentage of numbers repeated correctly when the vehicle is stationary be 100% at a certain frequency (ΔT) when the random numbers display one by one, and for a particular speed/curvature combination on the track, let the percentage be X% at the same display rate ΔT. The visual attention demand (VD) was estimated as follows:
where VD i is the visual attention demand at ith test track section. The VD provides a measure of an average percentage when a driver observing the roadway and traffic at a roadway section. As the driving scene becomes complex, the driver would need more visual attention demand to sense and analyze the information to control his/her driving, leading to less sampling of the secondary visual task. The VD is greater, which indicates the subject's visual demand for road traffic environment analyzing and driving behavior decision higher.
2.3. Experimental Design and Procedure
To examine attention demands and driving performance on different geometric alignment highway, a study with secondary visual task was conducted in the driving simulator. A one-way repeated measures design was used by presenting order of test sections. The virtual environment through which participants drove was a single visual database including rural sections. In the experiment, a 30 km stretch of two-lane rural highway with lane width equal to 3.50 m and a central barrier was simulated. The experimental route consisted of 20 geometric alignment sections (Figure 3, designed according to the highway engineering technique standard 2013 in China) with the section number from 1 to 20 and different curvature change rates (CCRs). The CCR was not only used to characterize the geometric alignments of experimental sections, but also the difficulty rates of driving tasks. The design speed of the highway was 60 km/h without traffic flows.
The experimental route consisted of 20 geometric alignment sections.
The primary task was driving at the speed of 60 km/h along the dotted marking on the central lane on KMRTDS. If drivers were traveling below or above 10 km/h, they would be audio reminded. If the driving behaviors were not “safe driving,” he/she were asked to drive the same track once more.
The primary hypothesis for VD testing is that the driver must make sure the divided attention to the secondary visual task under performing the safe driving (namely the primary visual task), which would not need any more visual attention.
Upon arriving in the laboratory, each subject driver was briefed on the requirements of the experiment and signed an informed consent document. Consistent with previous studies, each participant drove 20 min on a learning route. Then the subjects attended the stationary random number repetition test to 100% correction to get the ΔT. In the process of the subject's experiment, the ΔT was used to generate the random number at the same time rate. The objective was to eliminate the reaction differences of drives’ attention among subjects. After each trial, subjects filled out a Nasa-TLX workload rating form and a simulator sickness questionnaire. It took each driver 90 min to complete the trial.
2.4. Data Collection
In this study, we collected data on the driver's driving primary and secondary performance on each section. Drivers’ primary performances while driving the test track were recorded by the KMRTS. Although the simulator collected a number of parameters, we only processed the output data consisted of the time (with a gap of 0.02 s) and the corresponding coordinate (x, y, and z), longitudinal and lateral speeds and acceleration, steering angle, braking pedal displacement, and braking times. The lateral position was defined as the location of the vehicle's longitudinal axis relative to a longitudinal road reference system [20]. In our study, the lateral position corresponded to the distance (in m) from the vehicle centroid to the lane centerline. In the test trials, we selected 6 parameters as the driving performance indicators. Drivers’ secondary task performances were audio and screen recording. According to the subject's number repetition and the displaying number on a laptop randomly, the VD on each section was calculated artificially.
3. Results
Prior to any statistical analyses, a data screening was used to identify and exclude any response outliers that arose due to experimental equipment issues or participant failure to follow instructions during test trials, which were not expected in the experiment design. In this case, the experiment that the traveling average speeds are upon 75 km/h or below 50 km/h in a section was excluded from data analysis. In addition, the experiment that the lateral position went beyond the highway sideline was also excluded from data analysis. After the preliminary data screening process, data validation was conducted by hypothesis test using F test (ANOVAs) and Kruskal-Wallis nonparametric test to check for homogeneity of variance and average. A trend line analysis (TLA) was conducted to identify those behaviors with the greatest degree of variability. After the dataset validation, correlation analyses and regression analyses were conducted on driving performances for which interrelations were expected.
3.1. Driving Performance
Based on the results of TLA, the driving performance measures were selected to assess whether the driving performances are differing from each other on different CCRs sections. The independent variables are lateral position (m), longitudinal speed (km/h), longitudinal and lateral acceleration (m/s2), steering angle (deg), and braking times (times) in ANOVA tests, respectively. Statistical results and ANOVA results of driving performance measures (Table 1) indicated that the driving behaviors were significant differences on the overall test sections.
Statistical results and variance results of the single-factor analysis for 6 driving behaviors of the overall experimental section.
Additionally, results of Kruskal-Wallis nonparametric test for driving performance measures revealed their significance of the difference in different CCRs sections. The mean ranks of driving performance measures are significant difference (asymp. sig. < 0.01).
The driving performance profiles of all participants for all sections were examined in this study. Based on different CCRs sections, the trends of six driving behaviors were profiled. Figure 4 shows the different trends. The preliminary trend shows the lateral position increasing with the increase of CCRs (Figure 4(a)). The average and standard deviation of lateral position on section number 16 is the maximum. Because it is a sharp curve (R = 200 m, CCR = 5.26698), the subjects should keep the speed at 60 km/h, which leads not to maintaining driving on the central lane or adjusting traveling trajectory. There was a condition of acceleration on section number 4 led by an oval alignment, long traveling route, and easement curve. Therefore, the standard deviation of lane position on this section is large. The variety of longitudinal speed in section number 9 (CCR = 5.795114, Figure 4(b)) is the maximum, which is an S type section (R1 = 200 m, R2 = 360 m,). It was because this section required the subjects to keep tracking the S type route and keep adjusting the speeds to 60 km/h. In Figures 4(c) and 4(d), the trends of maximum variety of longitudinal and lateral acceleration were on sections number 16 and 9 because of their higher CCRs. Figure 4(e) showed the trends of steering angle. As a whole, the average and standard deviation of steering angle increased with the CCRs. Results in Figure 4(f) is breaking times trends, indicating how subjects controlled speed. More incidents of hard braking occurred in the condition where the accelerating contexts produced on sections number 5, 4, and 9. In order to control appropriate speed, drivers must make braking decisions and perform it.
Trends of driving behaviors with the corresponding CCRs: (a) lateral position, (b) longitudinal speed, (c) longitudinal acceleration, (d) lateral acceleration, (e) steering angel, and (f) braking times.
3.2. Development of Visual Attention Demand Prediction Models
The mean ranks of Kruskal-Wallis nonparametric test for subjects’ VD on 20 sections were significantly different, and the test statistics (Chi-Square = 39.577, Asymp. Sig. = 0.004) showed that the VD of different CCRs was different. There was a correlation between VD and CCR. Trends of VD on testing sections of different CCRs were shown in Figure 5. In general, we preliminary concluded that the VD was positively correlated with CCRs because of the increasing VD along with the growth of CCRs.
Trends of VD on testing sections with different CCRs.
To develop a prediction model of VD, we classified the CCRs by the same type alignment or similar value of CCR and calculated the corresponding average VD (shown in Table 2). The 20 test sections were classified by 11 kinds of sections. The average value was defined as the classified value.
Classified CCRs and classified VD on different alignment types sections.
Correlation results across the VD and CCRs indicated that drivers’ visual attention demand was closely correlated with the road geometric alignment, as expected.
A linear model and a quadratic model were obtained by the regression analysis (Figure 6). In the models, the dependent variable is classified VD and the independent variable is classified CCRs. The two models had good and similar goodness of fit. The model formula of linear regression is shown in formula (2), and quadratic model is shown in formula (3) as follows:
Fitting curves of classified CCRs and classified VD.
3.3. Interaction among Visual Attention Demand, Driving Behaviors, and CCRs
Correlation analyses were conducted to identify relations among visual attention demand, driving behaviors, and CCRs (Table 3). Nonparametric correlation analyses were used because of the fact that the observations of the measures were not all normal (correlation coefficients were noted by Spearman's ρ). From the correlation results across the VD to driving performances and CCRs, we concluded that there were close correlations among them. In the table, the parameters with noncorrelation were grey-shaded. These parameters were mainly longitudinal driving performances including longitudinal speed and acceleration. In addition, the braking times were noncorrelated with the VD.
The results of Non-parametric correlation analyses among VD, CCR and driving performance variables.
Correlation is significant at the 0.01 level (2-tailed)
correlation is significant at the 0.05 level (2-tailed).
4. Discussion
From an experiment design perspective, the manipulations of roadway alignments and visual attention demands were successful. Based on the results of driving performances, it appears that driving performance can be effectively influenced by the highway alignment. Previous studies have mainly paid close attention to speed [21, 22]. Although it is simple and direct, we cannot find out the influences on other driving behaviors. In this study, the driving performance including lateral and longitudinal behaviors showed trends variance with different alignment sections. The differences of lateral position, longitudinal speed, longitudinal and lateral acceleration, steering angle, and braking times conformed that highway alignment resulted in different driving performances. Although we should acknowledge that the differences’ results were induced partially by drivers’ driving style, the CCRs were the main influencing factor. In the present study, we found that these differences occurred when drivers were engaged in tracking the geometry alignment while driving on highways without traffic. Under the experimental conditions with the control of the speed of 60 km/h, the lateral position was influenced by the CCR, small radius sections, and acceleration condition. The CCR positively influenced the lateral speed and acceleration, steering wheel angle, and brake behavior, while significantly negatively influenced the longitudinal speed, not influencing the longitudinal acceleration. These results facilitated the study of the mechanism of CCR affecting driving behaviors.
As would be expected, with the increasing of the complicity of highway geometry alignments characterized by CCRs, the drivers’ visual attention demands increases too. In this study, the relation between classified CCRs and VD was confirmed. We classified the CCRs of 20 sections to 11 kinds of similar roadways according to the driving task difficulties and geometry alignments similarity. Two visual attention demand prediction models revealed their relationship. The goodness of fit of two models was 0.80 and 0.83, which indicated that we have selected and classified the typical geometric alignments. The drivers’ visual attention demands were sensitive to the difference of CCRs and the difficulty of driving tasks was also of high sensitivity. In the linear model, minimal visual attention demand on a straight section was 0.585 (when CCR = 0), meaning the average threshold of visual attention demands without dealing with the tracking. The average threshold in quadratic model was 0.596. There was not distinct difference. After constructing the model, we also validated the model with on-road test. We tested drivers’ visual attention demand on 7 sections in practice. The validation showed the effectiveness of the models.
As mentioned in Section 3.3, there were interactions among visual attention demands, driving behavior, and CCRs. In general, the greater the CCR of section was, the more the driver's visual attention demand was, and the more difficult the driving task was. The positive coefficient of the driving performance's indicator variable corresponding to CCRs indicated that the driving behaviors were influenced by the difficulty of driving task. The driving performance's indicator variable except the breaking times was positively correlated with the drivers’ visual attention demands. The reason was that the subjects automated on breaking control. The automation of breaking may help the driver, for example, in conditions where environmental demands are high [23]. In their interaction relationship, we found that the longitudinal speed did not correlate with breaking times, and the longitudinal acceleration did not correlate with the other parameters except for the VD. It was because the subjects couldn't control their speed of 60 km/h appropriately according to the experiment requirement, and then induced some problems of driving behaviors.
What is the influence of VD on driving performance? Although we found some correlations between VD and the driving performance, we couldn't find their quantitative relationship and model in common statistical analysis; we should study the mechanism of the VD influencing the driving behaviors.
5. Conclusions
In conclusion, the present study highlighted the relationship among the CCRs, the drivers’ visual attention demand, and the driving performances on a divided two-lane rural highways. The driving performance can be effectively influenced by the highway alignment. Significant differences were found in drivers’ behaviors while driving in different CCR sections. The present study concentrated on the model for the prediction of the drivers’ visual attention demands based on the measures of different curvature change rates. In general, the more complicated the highway geometry alignment is, the more the drivers’ visual attention demand will be occupied. The interactions among visual attention demands, driving behavior, and CCRs were also found.
The findings of these results may contribute to the highway's geometric alignment design and the drivers’ behavior model. In the future, to improve the prediction model, we should enlarge the subjects and the geometric alignment. Future research should consider how to explore the actual relationship between factors in more detail and more precisely, for example, the prediction model of driving performances influencing the drivers’ visual attention.
Conflict of Interests
The authors do not have any conflict of interests regarding the content of the paper.
Footnotes
Acknowledgments
The work described in this paper was partially funded by Natural Science Foundation of China under the contract number 71361016, and the applied fundamental science research project in Yunnan province under the contract number 2010CD043. The authors would like to thank their participants for the involvement in the experiment. The experiments were carried out at the Road Traffic Driving Simulation Laboratory at the Kunming University of Science and Technology.