category: literaturenote citekey: maoimprovinghazardperceptiondrivers2023 title: Improving hazard perception for drivers based on driving styles with visual alerts authors: "Mao, Yan; Wang, Xuan; He, Wu; Pan, Gaofeng" year: 2023 date: 2023-00-00 2023 doi: 10.1016/j.trf.2023.07.018 publication: "Transportation Research Part F: Traffic Psychology and Behaviour" url: "https://doi.org/10.1016/j.trf.2023.07.018" zotero_key: 6FNDTBQN zotero_storage: S2W936TA collections: transpordi_planeerimine folder: 001_artiklid firstAuthor: "Mao, Yan"

status: converted

Contents lists available at ScienceDirect

Transportation Research Part F: Psychology and Behaviour

journal homepage: www.elsevier.com/locate/trf

Improving hazard perception for drivers based on driving styles with visual alerts

Yan Mao ^a, Xuan Wang ^a, Wu He ^b,^∗, Gaofeng Pan ^c

^a School of Business, Sichuan Normal University, Chengdu, PR China
^b College of Movie and Media, Sichuan Normal University, Chengdu, PR China
^c The School of Cyberspace Science and Technology, Beijing Institute of Technology, Beijing, PR China

A R T I C L E I N F O A B S T R A C T

Keywords: Hazard perception Virtual reality Visual alerts Driving styles Driving safety

This paper aims to investigate the influence of driving styles on dangers by testing drivers' hazard perception with driving style variability in different scenarios and applying visual alerts to improve driver hazard perception. To achieve such goals, virtual reality systems are employed for simulations. In the experiment, we tested drivers' hazard perception with four driving styles (dangerous, angry, anxious, and cautious) in different hazard conditions and improved hazard perception through visual alert training. Participants' driving behavior and hazard perception were recorded simultaneously. The results showed that (1) Dangerous drivers have the weakest perception of danger, cautious drivers are the strongest, and anxious and angry drivers are similar. (2) Visual alerts can effectively improve drivers' hazard perception and help them mitigate hazards. (3). Visual alerts are more effective for dangerous drivers in improving hazard perception and mitigating hazards. The study offers important implications for road safety with visual alert training by improving hazard perception based on driving style.

1. Introduction

Today, traffic accidents significantly contribute to injuries and deaths (Castanier et al., 2013). Every year, more than 1.35 million people die in road traffic accidents worldwide, and another 50 million are injured (WHO, 2018). As hazard perception is considered one of the most effective skills in reducing driving collisions, the driver's ability to perceive hazards is crucial to reducing road accident casualties.

However, the research process needs to clarify risk and hazard perception. Risk perception is the process that mediates between risk-taking and risk utility. It is primarily a psychological process through which drivers. It is related to detecting environmental risks and the driver's self-assessment of the potential to avoid them. Hazard is defined as any situation or event in traffic that increases the probability of an accident (Renge, 1998). Therefore, hazard perception is similarly thought of as road situation awareness (Horswill & Mckenna, 2004), the ability to detect and recognize obvious or developing hazards in the driving environment (Mcknight & Mcknight, 2003). It is about observing hazards, assessing their extent, and avoiding them (Sheet, 2007). Hazard perception is generally more effective in avoiding hazards than risk perception. Therefore, hazard perception was chosen for this study.

Hazard perception is usually assessed through a test that requires learners to watch a series of video clips from the driver's perspective and press a button when they see a developing hazard. The quicker the driver reacts to the hazard, the higher the score

^* Corresponding author. E-mail address: wuhe83@163.com (W. He).

received, thus reflecting higher hazard perception skills and subsequent road safety (Crundall et al., 2021). Horswill demonstrated that hazard perception tests can measure potentially advanced skills related to driving (Horswill, 2017). Performance on such tests can even predict traffic crashes retrospectively and prospectively (Boufous et al., 2011, Horswill et al., 2015), significantly reducing collisions among young drivers (Wells et al., 2008). Therefore, it is relevant to study hazard perception because the driver's hazard perception ability is highly correlated with the risk of driving accidents. As driver experience increases, this ability increases, and then accident risk decreases (Ryu et al., 2020).

Furthermore, some researchers believe that the higher the driver's hazard perception, the higher the hazard reaction time, fixation reaction time, risk hit rate, and fixation probability (Cao et al., 2022). It is possible due to the driver's more comprehensive visual scan and broader spread of attention (Cao et al., 2022). However, drivers with a lower hazard perception are more likely to act in a risky manner (Wilde, 2010). For example, most young drivers underestimate risk and drive dangerously (e.g., by distracting themselves with mobile devices or speeding up) even after a crash (O'Brien et al., 2017). Drivers with higher levels of perception will change their behavior to reduce the risk to an acceptable level (O'Brien et al., 2017). In addition, studies have found that fear enhances drivers' perceptions of higher levels of risk. At the same time, anger triggers more optimistic risk perceptions and poor driving behavior, potentially increasing the likelihood of a collision (Mesken et al., 2007, Abdu et al., 2012). Although studies have shown that emotions can affect hazard perception, how drivers with different driving styles perceive it is unclear.

Driving styles refer to a driver's driving style or habitual driving pattern (Elander et al., 1993, Ishibashi, 2007). It is considered a critical factor in reflecting the internal state of driving and predicting driving behaviors (Fan et al., 2019). In studying driving styles, this paper selects the Multidimensional Driving Style Inventory (MDSI) for driving style classification. Because the MDSI is a valid and reliable indicator for assessing driver style (García et al., 2020, Long & Ruosong, 2019, Taubman-Ben-Ari & Skvirsky, 2016, Taubman-Ben-Ari et al., 2004). The MDSI classifies driving styles into The MDSI classifies four types: dangerous, angry, anxious, and cautious (Taubman-Ben-Ari et al., 2004). (1) Dangerous style refers to deliberate violations of safe driving norms and thrill-seeking while driving and is characterized by high speeds, racing, and illegal overtaking. (2) Anxious styles are associated with feelings of alertness and tension and the absence of practical relaxation activities while driving. (3) Anger style refers to the display of irritability and anger on the road, as well as hostile attitudes and behaviors, typified by a tendency to behave aggressively, such as swearing or flashing lights at other drivers. (4) Cautious style reflects an adaptive driving style that includes planning, paying attention to the road, being patient, courteous, and calm, and obeying traffic rules (Taubman-Ben-Ari et al., 2004).

Drivers with different driving styles have different risk perception abilities. Drivers with dangerous, angry, and anxious driving styles take more risks on the road than those with cautious styles (Taubman-Ben-Ari et al., 2004, Navon & Taubman-Ben-Ari, 2019). For example, drivers with an angry driving style have a higher risk of colliding with vehicles in the target lane when executing lane decisions (Huo et al., 2020). On the other hand, drivers with a cautious driving style usually plan their routes and drive carefully (Taubman-Ben-Ari et al., 2004). Most studies have focused on the effects of different driving styles on-road risk perception, but more research needs to be conducted on the hazard perception of drivers with different driving styles. Drivers' hazard perception is the ability to observe, quantify, and avoid hazards (Sheet, 2007). Different driving styles can differentiate the behavior of drivers in response to hazards. Therefore, hazard perception tests for drivers with different driving styles can provide further insights into the interaction between driving style and hazard perception.

Early research on hazard perception has focused on testing drivers with different driving experiences or safety situations using video or image formats (Jackson et al., 2009, Crundall, 2016). However, how can avoiding risk be improved despite understanding the driver's hazard perception? Some researchers have also proposed training to improve hazard perception accordingly. For example, road exposure increases driver awareness of dangerous road conditions (Borowsky et al., 2010, Upahita et al., 2018). Using forwardroad collision warning systems reduces rear-end accidents (Hajiseyedjavadi et al., 2018). In addition, personal computer-based and driving simulator training is an effective tool to improve the hazard awareness of young drivers and their visual scanning (Vlakveld et al., 2018). Moreover, nowadays, many scholars have studied the model by adding its remaining influences. By adding different visual scopes, examine their role in hazard perception, thus determining whether they enhance hazard perception (Crundall et al., 2021). Alternatively, improve the testing techniques for hazard perception (Ryu et al., 2020). Although this training can somewhat improve driver hazard perception, visual perception is the primary source of information for driving, and visual search behavior is necessary for hazard perception and driving performance (Huo et al., 2020). Therefore, visual alert training provides less driver distraction and more effective hazard avoidance (Hajiseyedjavadi et al., 2018). However, it is still being determined whether visual alert training can improve hazard perception for drivers with different driving styles.

So far, there are several approaches proposed for hazard perception testing. Firstly, the use of still images rather than dynamic clips (Huestegge et al., 2010, Scialfa et al., 2012). Secondly, the danger of using video recordings (Mckenna & Crick, 1994). The third is using computer-generated images (Malone & Bruenken, 2016). The fourth method is to use natural hazards (Horswill et al., 2015, Borowsky et al., 2010). Although these research methods effectively test hazard perception, hazard perception relies on behavioral and environmental subtleties not represented in the image world (Crundall et al., 2021). Therefore, to fill such a blank, this paper offers a meaningful way by using Immersive Virtual Environments (IVE) based on Virtual Reality (VR) technology.

VR is "a real or simulated environment in which the perceiver experiences a sense of remoteness" (Steuer, 2010). Because perceptual behaviors observed in virtual reality environments are qualitatively comparable to real-world behaviors (Kinateder et al., 2018). It has reasonable ecological validity and can evoke realistic human contingency behaviors. As well as being able to control and manipulate critical variables according to experimental design, reliable qualitative and quantitative behavioral data can be collected (Kinateder et al., 2014, Lin et al., 2020). Secondly, VR is beneficial for facilitating perception. The reason is that virtual environments can create a sense of presence, a state of consciousness, i.e., the (mental) sensation of being in a virtual environment (Slater & Wilbur, 1997). Moreover, this feeling helps conduct hazard perception tests because it makes the subject feel like he or she is driving down the road. Thirdly, VR technology is instrumental in driver rehabilitation. Because, unlike road tests, it allows the user to safely experience the driving situation by recreating the same traffic conditions, regardless of the administration time (Schultheis & Mourant, 2001).

Furthermore, applying the information obtained through PC-based hazard perception to actual driving situations is cognitively challenging (Agrawal et al., 2018). Hazard perception training using VR allows for better hazard perception training for drivers (Agrawal et al., 2018). Finally, given the obvious ethical and practical considerations of road hazard anticipation testing, the development of VR driving simulators provides a safer and more realistic environment for assessing drivers' hazard anticipation skills. Some countries even use VR hazard expectancy tests to obtain driving licenses. For example, Mangalore et al. (2019) evaluated whether VR headsets could assess drivers' hazard expectancy skills and compared young and middle-aged drivers when navigating in a VR HMD-based or stationary-based driving simulator. They demonstrated that VR headsets can effectively measure driving performance and address various research questions. Overall, it is safer and more effective to use VR for experiments (Mangalore et al., 2019).

In summary, the hazards of driving accidents for drivers are significant. To reduce it, the driver's hazard perception is critical (Ventsislavova et al., 2019). Although previous research has confirmed the importance of driver hazard perception in avoiding hazards, there is variability in the driving behavior exhibited by drivers with different driving styles. Furthermore, existing research on drivers with different driving styles needs to be improved. Therefore, understanding the hazard perceptions of drivers with different driving styles can help driver training providers tailor their driving simulations. Therefore, exploring the hazard perception of drivers with different driving styles is relevant. Secondly, although there has been research in recent years on improving driver hazard perception (Abele ¯ et al., 2019), it is unclear whether using virtual reality to form visual warnings can improve drivers' hazard perception with different driving styles. It also needs to be determined how effectively visual warnings improve hazard perception for drivers with different driving styles. Therefore, these research gaps have driven the present study.

Therefore, this paper aims to investigate the hazard perception of drivers with different driving styles using virtual reality technology. Secondly, to investigate whether visual cues can improve drivers' hazard perception. In the following sections, it will be explored in detail.

2. Related work

2.1. Hazard scenarios

Hazard-aware test-driving scenarios are also crucial. Researchers have classified hazards into four main categories: visible materialized hazards, visible non-materialized hazards, implicit materialized hazards, and implicit non-materialized hazards (Kahana-Levy et al., 2019). The focus of this study is on materialized hazards. These planted hazards represent different avoidance potentials and situational complexities. A previous study has shown that risk perception is influenced by two main factors: information about potential hazards and the ability to prevent these hazards from becoming actual accidents (Kahana-Levy et al., 2019). Therefore, eight hazard scenarios were tested during the training and formal phases. The main ones include the Sudden appearance of pedestrians, Trucks parked in pedestrian crossings, Vehicles suddenly changing lanes, Signal changes, Right-turning vehicles, Sudden brakes, Pedestrians by the roadside, and the Sudden appearance of a motorcycle.

Sudden appearance of pedestrians: A vehicle may not brake in time and collide with a pedestrian (Zhou et al., 2022).

Trucks parked in front of pedestrian crossings: Trucks parked in front of pedestrian crossings can be in the way of the driver's scanning view. Trucks parked in front of pedestrian crossings can be in the way of the driver's scanning view, preventing the driver from observing potential hazards (Chen et al., 2013).

Vehicles suddenly changing lanes: Drivers often perform lane-changing driving tasks on the road to reach a planned destination (forced lane change) or to achieve desired driving conditions (unexpected lane change) (Chen et al., 2014). Although lane changing reduces the speed of neighboring road users, it can disrupt the surrounding traffic flow. As a result, lane changes are often reported as the cause of collisions (e.g., tailbacks and sideswipes) (Zheng et al., 2011a, 2011b, Li & Sun, 2017).

Signal change: Intersections are high-risk locations where traffic accidents occur frequently. The sudden appearance of a pedestrian under a Moreover, drivers have a short reaction time when pedestrians appear, leading to traffic. Moreover, drivers react quickly when pedestrians appear, leading to traffic conflicts and collisions due to wrong decisions (Lu et al., 2015). In contrast, traffic signals at intersections have the same impact on drivers' driving decisions as stop signs. In particular, the decision drivers make at the start of a yellow traffic light is often crucial, as wrong decisions always cause traffic conflicts and collisions (Haque et al., 2016).

Right-turning vehicles: Wrong turns, U-turns, or lane changes are the leading causes of rear-end and tailgating accidents (Ya et al., 2020).

Sudden braking can easily cause rear-end and tailgating accidents (Ali & Haque, 2023), resulting in a collision between two vehicles that can lead to injury or death.

Pedestrians by the roadside: Like pedestrians crossing the road.

The sudden appearance of a motorcycle: Not only is the motorcycle's driver vulnerable to injury, but blindness (Chen et al., 2023) can cause a collision between two vehicles in front and behind, resulting in multiple casualties.

2.2. Response latency (RL)

Response latency (RL), sometimes called reaction time, is a performance metric in many driving studies (Upahita et al., 2018). RL is usually calculated as the time interval between the first appearance of a stimulus on the screen and the subject's response to the inspiration and is represented in traditional hazard perception tests by pressing a button. Researchers often divide the response latency into two components: time to first fixation (TFF) and response time (RT) (Hajiseyedjavadi et al., 2018). Upahita et al. (2018) further modified the TFF and RT components to analyze response time, and the present study adopted that approach for the investigation to divide RL into three pieces. TFF represents the time from the first appearance of the hazard until the eye's gaze reaches the object of interest. The TFF represents the time from the first appearance of the danger to the time when the eye gazes at the object of interest, i.e., the dangerous object. When humans respond to a visual stimulus, it takes some time for the neurons to process the inspiration and send instructions to the muscles to move toward the object, which delays detection. Whereas the TFF component can be further decomposed into two pieces, lag time (LT) and fixation time (TF), which are classified as the detection component of the response time, respectively, LT represents the delay in the start of eye movement towards the hazard. It means the time from the first appearance of the object in the driver's view until the gaze point starts to move in the direction of the target object. TF is the time required to move the eye from the previous gaze point to reach the target object. It is referenced from the time the initial gaze point begins to move in the direction of the target hazard object until the eye is on the target object and then continues to follow that object.

The second component of response time is the decision-making component, called PIEV time. PIEV is often considered the component that influences a person's response. PIEV stands for perception (P)—time to identify the object; intelligence (I)—time to understand the risk imposed by the entity; emotion (E)—time to decide to act; and volition (V)—time to apply the decision to the actual action. It is often widely used to describe the structure describing the driver's perception and reaction processes, explaining that perception-reaction time can be decomposed into these four components (Upahita et al., 2018). In the present study, PIEV time was defined as the time from the first gaze at the target hazardous object to the time when the participant applied a response action (i.e., stepping on the brakes), which reflected the speed at which the participant responded to the stimulus. During the PIEV time, the driver goes through a process that involves viewing the object, understanding its meaning, deciding, and applying it to the action. These two components, detection, and decision-making, comprise the total response time, known as response latency (RL).

3. Method and materials

3.1. Experiment one

3.1.1. Participants

Eighty undergraduate or postgraduate students were recruited to participate in this study via online recruitment from universities in Chengdu, Sichuan, China. Drivers must have a valid driver's license and at least one year of driving experience. Participants had normal or corrected-to-normal vision and hearing. Participants were required to sign a written informed consent form before the start of the experiment, complete a pre-driving questionnaire that included questions related to demographics, driving history, and driving behavior, and familiarize themselves with the driving equipment. Upon completion of the experiment, each participant received CNY 20.00.

The Sichuan Normal University Ethics Committee approved the study protocol under the International Declaration of Helsinki. Study participants were given verbal consent, and participation was based solely on their wishes. Participants were also informed of the purpose of the study and its procedures. In addition, participants were informed of their right to withdraw from the study.

3.1.2. Apparatus

The study was conducted in a virtual reality lab. We used the HTC VIVE head-mounted display (HMD) VR system, which uses SteamVR positioning technology and allows a 360° view of the display. The field of view was 110°, and the display's resolution was 1080 (horizontal) x 1200 (vertical) pixels for each eye.

The experiment was run on a Dell Precision T7800 workstation with an Intel Xeon E5-2603 processor and a GBGTX1080 graphics card. We used 3D Studio Max software to model and render the IVE (immersive virtual environment) and then imported it into the Unity3D game engine's platform. We also used the particle system visualization process in Unity 3D.

During the experiment, participants remained seated and used the steering wheel to interact with the IVE. Specifically, they manipulated the steering wheel to move through the IVE, either forward, backward, left, or correct, while keeping their bodies still.

Lane maintenance measures are essential to the study. Not only does it ensure that the experiment is carried out correctly and that participants can participate in it more safely, but it also makes the results more relevant and valid. Therefore, the VR system is used to identify information about the vehicle along the lane it travels, determine whether the vehicle deviates by identifying vehicle information and lateral deviations, and, if so, whether it needs to be corrected.

3.1.3. Experimental scenarios

This experiment uses a VR system for hazard perception testing, which lasts approximately 10 minutes and is conducted from the driver's point of view. Participants were all exposed to the same stimuli in the same way. The experiment consisted of five hazard scenarios that required the driver to respond quickly in the face of danger. The response is defined as any behavior that requires the driver to take corrective action (e.g., applying the brakes, using the steering wheel to take evasive action) to avoid a hazard (e.g., a potential collision) (Ryu et al., 2020).

The hazardous scenarios include pedestrians suddenly appearing in view, trucks stopped in front of pedestrian crossings, sudden lane changes, signal switches, right-turning vehicles, vehicles braking ahead, crossing pedestrians, and crossing motorbikes, all of which are presented in the IVE environment (as shown in Fig. 1 and Table 1).

Table 1 Experimental scenarios.

Dangerous scenes	Description
Hazard 1: Sudden appearance of pedestrians	The test vehicle was traveling along one of the main roads when, in front of the test vehicle, it started to cross a pedestrian visible at the edge of the pedestrian pavement.
Hazard 2: Trucks parked in pedestrian crossings	While the test car was driving along the left lane, a truck stopped in the left lane of the pedestrian crossing(three lanes).
Hazard 3: Vehicles suddenly changing lanes	On the driving road, a truck comes from the left side of the test vehicle and merges with the test vehicle's path.
Hazard 4: Signal change	The traffic lights change from green to yellow as the test vehicle approaches a junction on the main road.
Hazard 5: Right-turning vehicles	As the test vehicle approaches a junction on the main road, a vehicle comes out of the left-hand side of the path and then turns right across the test vehicle's path.
Hazard 6: Sudden brake	The test vehicle was traveling along the main road. The vehicle in front of the test vehicle suddenly slowed down.
Hazard 7: Pedestrians by the roadside	The test vehicle was traveling along the main road. The driver at the right of the test vehicle suddenly opened the door and got out of the vehicle, preparing to cross the road.
Hazard 8: Sudden appearance of a motorbike	The test vehicle is traveling along the main road. Suddenly, a motorcycle is inserted in front of it.

Fig. 1. Hazard scenarios. (a) Hazard 1: Sudden appearance of pedestrians; (b) Hazard 2: Trucks parked in pedestrian crossings; (c) Hazard 3: Vehicles suddenly changing lanes; (d) Hazard 4: Signal change; (e) Hazard 5: Right-turning vehicles; (f) Hazard 6: Sudden brake; (g) Hazard 7: Pedestrians by the roadside; (h) Hazard 8: Sudden appearance of a motorbike.

3.1.4. Multidimensional Driving Style Inventory (MDSI)

Drivers complete a Likert scale based on statements about their feelings, thoughts, and behaviors while driving. The scale has a score of 6, ranging from 1 (not at all) to 6 (very much). As the Cronbach's alpha for the four dimensions was reasonable (0.81 for dangerous driving, 0.81 for anxious driving, 0.77 for angry driving, and 0.70 for cautious driving), each driver's responses to the relevant scales were averaged to produce scores for each of the four driving styles. For each of the four driving styles, a higher score indicated a higher level of that style. The final scores were 18 for the dangerous driving style (ten males and eight females), 21 for the angry driving style (Twelve males and nine females), 19 for the anxious driving style (eight males and Eleven females), and 22 for the cautious driving style (Eleven males and Eleven females).

Furthermore, Drivers with a cautious style are usually easy to distinguish, as they generally show compliance with traffic rules. Anxious drivers, on the other hand, are usually more tense and have no relaxation activities. For the danger and anger-style drivers, we used the findings proposed by Van Huysduynen to differentiate: dangerous driving style is associated with mean speed and standard deviation of lateral position (e.g., high speed, illegal overtaking), while anger driving style is positively associated with driving speed and negatively associated with forward distance (e.g., swearing behavior, flashing lights) (Van Huysduynen et al., 2018).

3.1.5. Design reliability and validity item

The study used a between-group design to investigate the hazard perception of drivers with different driving styles (dangerous, anxious, angry, and cautious) when faced with different hazard scenarios (H1–H8). The three main aspects of hazard perception were speed, reaction latency time, and Response rate. The MDSI questionnaire was used to classify these drivers into four driving styles, each experiencing five hazard scenarios. This paper uses speed, steering wheel departure angle, the use of brakes or not, and hazard response time as indicators of the driver's ability to mitigate hazards.

Fig. 2. Experimental procedure.

Speed choice scores measure subjects' tendency to avoid or take risks (Renge, 1998). When a driver chooses a lower speed, he or she avoids risk. The steering wheel departure angle and the use of the brake or not can intuitively avoid the occurrence of the risk.

Response latency (RL), sometimes called reaction time, has been used as a performance metric in many driving studies (Upahita et al., 2018). RL is typically calculated as the time interval between the first appearance of a stimulus on the screen and the subject's response to the stimulus. The traditional HPT (hazard perception test) measures it by pressing a button. Upahita et al. (2018) defines RL as

$$RL = LT + TF + PIEV \tag{1}$$

Therefore, this definition was also used in this study. The lag time (LT) indicates the time from the onset of the hazard on the screen until the eye starts to move in the direction of the target hazard object. The gaze time (TF) is the duration from when the participant starts moving the gaze point to the time the participant looks at the target hazard. It is a mechanical process of eye-gaze point movement. It is the shortest component of RL. Perceptual-intellectual-emotional-volitional (PIEV) time represents the time interval required for a participant to perceive the hazard he sees, decide, and implement that decision into action. A smaller RL indicates that the driver is better able to perceive the hazard.

3.1.6. Procedure

First, participants were tested in the laboratory. The researchers provided written instructions, and the participants had to sign a written consent form detailing their right to withdraw without cause. They were also asked to complete a pre-driving questionnaire. These included questions related to demographics, driving history, driving behavior, familiarization with the driving equipment, and the experimental procedure, as shown in Fig. 2.

Next, participants were asked to complete the MDSI questionnaire. Once the grouping was complete, tasks were assigned to participants. The first, driving training, required participants to familiarize themselves with the VR equipment and the virtual driving environment. Secondly, a hazard perception test was conducted after the training was completed, and they were informed of the need to react quickly to hazards when they occurred. The hazard test consisted of eight scenarios: sudden pedestrians suddenly appearing in view, a truck stopped before a pedestrian crossing, a vehicle suddenly changing lanes, a signal switch, a right-turning vehicle, a sudden brake, Pedestrians by the roadside, and the sudden appearance of a motorcycle.

Upon completion of the test, participants are given an evaluation questionnaire. The questionnaire includes feedback and thoughts on the driving experience and the IVE environment, as well as an assessment of the effect of the NASATLX questionnaire on the participant's motion sickness.

The participant is informed during driving: "Drive as usual. The environment will contain five hazards. When you observe a hazard in the environment, you must act by changing speed, adjusting the steering wheel, or braking to avoid a collision. Ensure you try to keep driving generally at all other times and only make changes when hazards occur so we can use your data properly."

3.2. Results

The study was explored using empirical analysis. Firstly, the effects of demographic variables (e.g., gender, age) on driving style and risk response time were examined using independent sample t-tests. Secondly, the effects of driving style, speed, and response latency time were examined using an analysis of variance. Finally, after controlling for demographic variables, a binary logistic regression analysis was conducted to test the effects of response latency time and driving style on risk perception.

Table 2 Driving style score according to gender.

Driving style	MDSI factors	Men	Women
Angry driving style	Risky and Angry
	Mean	2.63	2.13
	S.D.	1.08	0.94
Dangerous driving style	High-velocity and Distress reduction
	Mean	2.71	2.26
	S.D.	0.78	0.77
Anxious driving style	Dissociative and Anxious
	Mean	2.16	2.25
	S.D.	0.66	0.64
Cautious driving style	Patient and Careful
	Mean	4.42	4.63
	S.D.	0.75	0.68

Fig. 3. Speed.

3.2.1. Driving styles

The driver completes a Likert scale based on statements of feelings, thoughts, and behaviors while driving. The scale has a score of 6, ranging from 1 (not at all) to 6 (very much). Cronbach's alpha had a reasonable four dimensions (0.81 for dangerous driving, 0.81 for anxious driving, 0.77 for angry driving, and 0.70 for cautious driving), and each driver's responses to the relevant scales were averaged to produce scores for each of the four driving styles. For each of the four driving styles, a higher score indicated a higher level of that style. Thus, driving styles were classified according to gender differences, and differences in driving styles and gender were investigated by t-tests, yielding the resulting means and SDs shown in Table 2. The analysis showed significant gender differences, which indicates that males scored higher on dangerous and angry driving styles than females. Females scored higher on anxious and cautious driving styles than males.

3.2.2. Hazard perception under driving styles

The ANOVA on driver speed showed a significant effect of driving style and speed F (3,79) =2.57, p<0.05, 2=0.67. The slowest driving speed was found for cautious drivers and the highest for dangerous drivers (Fig. 3). However, there was no main effect for either hazard scenario, as F (3,79) =1.69, p<0.05, 2=0.15.

3.2.3. Response latency (RL)

The initial data checks show that the response time (RL) and components are not normally distributed. Therefore, we conducted a log transformation of the raw data for the parameters. Table 3 shows the mean of RL and its three components. Then, a oneway ANOVA was performed on RL and its components for all risk factors, with the group as the independent variable. Table 4 summarizes the statistical tests' results and indicates the parameters' differences between participant groups. It is noteworthy that the conservative group usually has the fastest response.

All components of RL were obtained from the data. The RL values were obtained by summing the lag time (LT), the fixation time (TF), and the PIEV time. Fig. 4 shows the mean and standard error of the RL, and the conservative group performed the best of the four groups of participants. Cautious drivers were more observant of road information and could quickly react to hazards. Moreover,

Fig. 4. Mitigation percentage.

Table 3 Average value of RL and its components. The result made by the driver when confronted with hazards. The main components include the driver's reaction time to the hazard and the four components of the reaction time.

Component	Groupa	H1b	H2b	H3b	H4b	H5b	H6b	H7b	H8b
RL(s)c	Cautious	1.513	1.578	1.552	1.566	1.543	1.502	1.589	1.573
	Dangerous	2.117	2.064	1.769	1.643	1.722	2.412	2.359	2.188
	Anxious	2.325	1.749	2.106	2.140	2.118	2.143	2.058	1.844
	Angry	2.394	2.356	2.046	2.848	2.674	2.325	2.359	2.125
LT(s)d	Cautious	0.478	0.428	0.489	0.508	0.496	0.468	0.476	0.437
	Dangerous	0.765	0.578	0.712	0.636	0.648	0.899	0.642	0.687
	Anxious	0.718	0.479	0.686	0.735	0.712	0.778	0.619	0.714
	Angry	0.889	0.668	0.799	0.838	0.803	0.718	0.698	0.734
TF(s)e	Cautious	0.261	0.382	0.382	0.256	0.287	0.262	0.380	0.258
	Dangerous	0.336	0.356	0.282	0.259	0.321	0.351	0.331	0.361
	Anxious	0.389	0.301	0.358	0.332	0.342	0.356	0.343	0.257
	Angry	0.352	0.329	0.368	0.464	0.367	0.398	0.301	0.328
PIEV(s)f	Cautious	0.774	0.768	0.681	0.802	0.760	0.772	0.673	0.758
	Dangerous	1.016	1.130	0.775	0.748	0.753	1.162	0.862	1.258
	Anxious	1.218	0.969	1.062	1.073	1.064	1.208	0.683	0.921
	Angry	1.153	1.359	0.879	1.546	1.504	1.106	1.058	0.993

^a Four groups of participants were divided using the MDSI questionnaire.

RL in the conservative group was significantly faster on hazard H1 than in the other groups, suggesting that the conservative group reacted quickly by giving way to pedestrians in that hazard scenario. At the same time, the other groups waited until the hazardous situation developed further.

Meanwhile, the results also showed that the group with the most dangerous driving style generally performed the worst and reacted more slowly to dangerous situations than the more cautious group. It shows that dangerous driving reduces vehicle control, and the driver's ability to estimate the situation and respond is also reduced. As expected, the hazard group took more time to react than the caution group. The hazard group did not usually make hazard mitigation decisions.

Of course, the anger and anxiety groups were intermediate in their perception of danger, and the anxiety group was more capable of perceiving danger than the anger group. However, they had lower levels of aggressiveness compared with the danger group. Therefore, the anger group had a lower hazard perception capacity than the anxiety group and a higher capacity than the danger group.

The RL was further broken down into three components, discussed below.

^b Dangerous scenarios.

^c Time for participants to respond to the hazard.

^d The lag time (LT) indicates the delay before the point of gaze moves when a hazardous situation is detected.

^e Time-to-fixation (TF) is the duration from when the participant started moving the fixation point to the time that the participant fixated on the target danger point.

^f Perception-Intellection-Emotion-Volition (PIEV) time represents the time interval that a participant perceives the hazard he sees, decides, and implements into action.

Table 4 Result of statistical test for RL, LT, TF and PIEV time.

| Hazard | Variable | P-value | ANOVA | | Post-hoc | Tukey HSD | | |----------|------------------|-------------------------|-------------------------|-------------------------|----------------------------------------------------------------|-------------------------|--| | | | | F value | P-value | Participant group | P-value | | | Hazard 1 | RL | 0.258 | 6.483 | 0.001 | Cautious < Dangerous
Cautious < Anxious
Cautious < Angry | 0.002
0.033
0.003 | | | | LT
TF
PIEV | 0.176
0.476
0.589 | 0.654
0.966
2.531 | 0.583
0.413
0.064 | Cautious < Dangerous | 0.085 | | | Hazard 2 | RL
LT | 0.113
0.470 | 3.089
1.354 | 0.033
0.264 | Cautious < Dangerous | 0.028 | | | | TF
PIEV | 0.256
0.634 | 0.627
2.548 | 0.597
0.063 | Cautious < Dangerous | 0.052 | | | Hazard 3 | RL
LT
TF | 0.722
0.934
0.980 | 2.653
0.614
1.047 | 0.057
0.609
0.372 | | | | | | PIEV | 0.773 | 2.662 | 0.057 | Cautious < Dangerous | 0.102 | | | Hazard 4 | RL | 0.189 | 9.090 | 0.000 | Cautious < Dangerous
Cautious < Angry | 0.000
0.001 | | | | LT
TF | 0.290
0.426 | 1.776
1.612 | 0.159
0.190 | | | | | | PIEV | 0.091 | 4.231 | 0.018 | Cautious < Dangerous | 0.004 | | | Hazard 5 | RL | 0.203 | 2.552 | 0.062 | Cautious < Dangerous
Cautious < Anxious
Cautious < Angry | 0.052
0.059
0.057 | | | | LT
TF
PIEV | 0.286
0.434
0.459 | 1.356
1.045
3.196 | 0.123
0.414
0.053 | | | | | Hazard 6 | RL
LT
TF | 0.557
0.387
0.167 | 1.325
0.338
1.399 | 0.035
0.278
0.272 | | | | | | PIEV | 0.798 | 3.084 | 0.058 | Cautious < Dangerous | 0.065 | | | Hazard 7 | RL
LT
TF | 0.982
0.203
0.424 | 2.037
0.472
1.355 | 0.145
0.628
0.272 | | | | | | PIEV | 0.643 | 3.460 | 0.031 | Cautious < Dangerous | 0.038 | | | Hazard 8 | RL | 0.062 | 5.747 | 0.006 | Cautious < Dangerous
Cautious < Anxious
Cautious < Angry | 0.015
0.016
0.021 | | | | LT
TF
PIEV | 0.047
0.646
0.017 | 0.289
1.154
2.581 | 0.748
0.327
0.089 | | | |

3.2.4. Lag time (LT)

The lag time (LT) indicates the delay before the point of gaze begins to move when a hazardous situation is detected. This study measures LT from the start of the hazard to the movement toward the target hazard object.

The cautious group had the fastest LT speed of the four hazards. It indicates that cautious driving increases the speed of detection for drivers. Cautious drivers may be more sensitive to changes in their surroundings when more information about the road is observed. Therefore, a hazardous object triggers the cautious driver's eye movements more quickly.

At the same time, there was no significant difference in LT between the anger and anxiety group groups. Moreover, Dangerous drivers have the lowest detection speed. Because dangerous drivers usually perceive themselves as highly skilled and prefer exciting driving styles, the perception of danger is usually low.

In summary, results show that drivers with different driving styles have different hazard ratings. Cautious drivers are more sensitive to hazards on the road. Meanwhile, drivers in angry and anxious groups may only react to dangerous situations when the danger is more prominent. Dangerous drivers usually ignore hazards.

3.2.5. Time-to-fixation (TF)

Time-to-fixation (TF) is the duration from when the participant started moving the fixation point to the time that the participant fixated on the target danger point. As shown in Table 4, there were no significant effects on TF values. It is because TF is closely related to a physiological mechanism of the eye. Therefore, given that all participants were in a similar age range, their physiological conditions were likely to be more similar, resulting in similar TF.

3.2.6. PIEV time

Perception-Intellection-Emotion-Volition (PIEV) time represents the time interval a participant perceives the hazard he sees, decides, and implements into action. A longer mean value for PIEV indicates that the decision process takes more time in the hazard mitigation process.

Fig. 6 shows the mean and standard error of PIEV. Overall, the cautious group had faster PIEV times than the other three groups, indicating that cautious driving has better performance for driving decisions. Thus, cautious drivers could make faster decisions and translate them into action. The ANOVA found a significant difference between the dangerous and cautious groups in H5. It showed the effect of hazardous drivers and cautious drivers on hazard reactivity. The H4 has a shorter crash time, which requires a faster response from the driver to avoid a crash. So, there is an effect of dangerous and cautious driving under intense pressure. The cautious driver can make quicker decisions. However, dangerous drivers cannot make decisions as quickly as cautious drivers because they ignore hazard information. As a result, dangerous drivers wait longer until they are closer to the hazard before evading action, resulting in longer PIEV times.

3.2.7. Correlation analysis

Firstly, participants' mean age is 22.1 years, SD = 2.6. The associations between drivers' driving styles, speed, Mitigation percentage, and hazard reaction times are shown in Table 5.

Table 5 Associations among driving style, speed, mitigation percentage and RL (n = 80).

Variable	Dangerous driving style	Anxious driving style	Angry driving style	Cautious driving style
Gender	-0.300∗	0.377∗∗	-0.293∗	0.115
Age	-0.197	-0.312∗∗	-0.018	0.240∗
RL	0.15∗∗	0.13∗∗	0.17∗∗	-0.25∗∗
Speed	-0.14∗∗	-0.09∗	-0.13∗∗	0.15∗∗
Mitigation percentage	-0.07∗∗	-0.06∗	-0.14∗∗	0.17∗∗

^*p <0.05.

^**p <0.01.

Fig. 7. Visual alerts.

Table 5 shows that anxiety, anger, and danger are positively associated with RL and negatively associated with reaction speed. Cautious driving style was positively associated with reaction rate, negatively associated with RL. In addition, anxiety, anger, and danger were negatively correlated with the Mitigation percentage, and cautiousness was positively correlated with the Mitigation percentage. The results also demonstrate that cautious drivers have the best hazard perception. Similar patterns between variables persisted after controlling for participants' gender and age.

3.3. Experiment two

3.3.1. Participants

After participants completed Experiment 1, they rested for 2 hours and trained for Experiment 2.

3.3.2. Experimental apparatus and materials

Experiment 2 differs from Experiment 1 in that it uses eye-tracking technology to provide visual alerts before the driver perceives the hazard, while the experimental equipment, scenario, and group classification are the same as in Experiment 1.

The visual alert is a simple representation of the road condition, showing only the road and potential hazards, excluding other features on the road, to prevent visual confusion and enhance perception. The visual image has a black background, and the hazard is yellow. The driver's seat is in front of the center of the screen, approximately 60 cm from the left edge (as shown in Fig. 7). The visual alert time was 4 s, considered valid (Hajiseyedjavadi et al., 2018).

3.3.3. Experimental design and procedure

Different driving styles drivers were tested again. A visual alert system validated the test before the driver was about to encounter danger and collected driving speed, whether they were watching, and the watch duration.

Eighty subjects were recruited for the experiment, which proceeded as follows.

First, participants were tested in the laboratory. The researchers provided written instructions, and participants had to sign a written consent form detailing the right to withdraw without cause at any time. They were also asked to complete a pre-driving questionnaire, which included questions related to demographics, driving history and behavior, and the driving equipment's familiarity.

Next, tasks were assigned to participants. The first, driving training, required participants to become familiar with the VR device and the virtual driving environment. Secondly, additional instructions were provided to the participants. The participants were informed that a visual alert would be presented on the central screen in case of a potential collision risk.

Upon completion of the test, participants are given an assessment questionnaire. The questionnaire includes feedback and thoughts on the driving experience and the IVE environment, as well as an assessment of the effect of the tool on the participant's motion sickness using the NASATLX questionnaire.

Table 6 Visual alerts.

Hazard	Unattended	Duration of gaze	Percentage of gaze	Latent	Failed
Hazard 1	3.31	0.64	22.4%	0.73	0.11
Hazard 2	2.83	0.57	26.6%	0.38	0.13
Hazard 3	4.15	0.67	24.7%	0.29	0.07
Hazard 4	1.86	0.59	33.5%	0.44	0.06
Hazard 5	1.92	0.57	40.3%	0.53	0.18
Hazard 6	1.52	0.66	23.1%	0.34	0.04
Hazard 7	1.50	0.47	24.1%	0.28	0.00
Hazard 8	2.01	0.60	33.3%	0.46	0.09

3.4. Result

3.4.1. Visual alerts

To better observe the effect of visual alerts on driver hazard perception, this paper calculates the average number of times that a driver views a warning, the average duration of the view, the average view delay (the time interval between the start of the warning and the first view of the warning), and the proportion of drivers who do not view the warning within the time provided. The results are presented in Table 6 , where warnings observed for less than 100 ms were excluded from the analysis because this time was too short for drivers to perceive, understand, and predict what the warning implied. A failure to look at the warning before it disappeared was considered a failure to obtain information from it. Overall, drivers viewed the warnings an average of 2.29 times, with a duration of 0.589 s. The time interval between the start of the warning and the first sighting was 0.448 s. The drivers successfully understood 88.9% of the warnings.

Warning gaze counts followed a Poisson distribution, so a mixed effects Poisson model was used to determine whether the driving style and scenario affected gaze counts. Results showed that the main effects of both [F (3,79) = 9.14, p<0.001] and the scenario [F (3,79) = 10.74, p<0.001] were significant, while their interaction was not [F (3,79) = 0.240, p>0.05]. We analyzed the dependent variables on a scenario-by-scenario basis. In Scenarios 1 and 4, drivers viewed warning messages more frequently, but in Scenarios 2, 3, and 5, drivers rarely considered warning messages (shown in Table 6). This result can be explained by the fact that in Scenarios 1 and 4, the driver needs to stop before approaching a hazard, but in Scenarios 2, 3, and 5, the driver does not need to stop to avoid the hazard and therefore spends relatively little time identifying the potential hazard.

There were no significant main effects [Group: F (3,79) = 0.43, p = 0.63; Scenario: F (3,79) = 0.62, p = 0.76] or interaction effects [F (3,79) = 1.13, p = 0.41] for viewing duration (Table 6). It indicates that the timing of the warning or driving conditions did not affect the viewing duration. A similar analysis was conducted for the percentage of views variable. The main effects of Group [F(3,79) = 6.47, p = 0.007] and Scenario [F (3,79) =10.17£¬p<0.0001] were both significant, but their interaction was not. [F (3,79) = 1.26, p = 0.24].

There was no significant effect of delay time for group [F (3,79) = 2.33, p = 0.14] or scene [F (3,79) = 1.47, p = 0.23], and the interaction was not significant [F (3,79) = 0.73, p = 0.87]. The mean viewing onset latency was 0.448 s, indicating that these warnings attract drivers' attention. Analysis of the likelihood of drivers viewing and successfully perceiving the message yielded an average of approximately 88.9% of warnings being successfully perceived by drivers.

3.4.2. Speed

The average speed of the subjects in the warning group (Experiment 2) was compared to the average speed of the control group (Experiment 1) separately, as shown in Fig. 8. It was found that there was a significant change in the speed of the drivers who gave the warning 4 s before the appearance of the hazard. Drivers who received the warning maintained a significantly lower speed as they approached the hazard than those in the control group. There was also a significant reduction in driver speed in all five hazard scenarios.

4. Discussion

4.1. Demographic variable

The study's results showed gender differences in drivers' driving styles, but no hazard perception was affected (Table 5). In other words, male drivers had a higher risk and an angry driving style, while female drivers tended to have an anxious and cautious style. It is consistent with previous research findings (Taubman-Ben-Ari et al., 2004). Again, age was a factor, with older age being associated with lower endorsements of the dangerous and angry style and higher reports of the cautious style. It may be because both male and younger drivers prefer thrill-seeking driving, while females tend to be low in thrill-seeking as well as low in injury. In addition, the results also indicated that male subjects were less able to perceive danger than female subjects. The male drivers may have been more confident in their driving skills. It is like previous research showing that male drivers choose faster speeds and are more inclined to be optimistic (Sun et al., 2021). Furthermore, gender effects are usually found alongside the effects of age. However, in this study, no effect of age was found in the analysis for RL. It might result from the need for a wide range of ages for this study.

Fig. 8. Speed comparison.

4.2. Hazard perception

The primary purpose of this paper is to explore the hazard perception abilities of drivers with different driving styles. The study's results suggest that cautious driving may mitigate hazards. In contrast, dangerous driving increases the probability of hazards occurring, while angry and anxious driving has no substantial hazard mitigation effect.

The result can be explained by the fact that, firstly, in hazard scenarios, choosing to turn the steering wheel and brake for mitigation behavior is practical. The collected data shows that a lower percentage of dangerous drivers (35.5%), and a higher percentage of cautious drivers (83.5%), adopted the mitigation behavior. Angry (59.33%) and anxious (53.55%) drivers used a similar proportion (Fig. 5). The result can be explained by the fact that dangerous drivers are confident about their driving behavior and usually make impulsive driving decisions. Therefore, they usually avoid dangerous mitigation behaviors when faced with dangerous scenarios. Therefore, when a hazard arises, they can take the right actions to avoid it.

Secondly, comparing drivers' performance with different driving styles can indicate their level of hazard perception. The response time (RL) showed that the cautious group responded more quickly than the dangerous group, while the angry and anxious groups did not differ significantly. The results suggest that careful driving can mitigate hazards. In contrast, dangerous driving increases the probability of hazard occurrence, while angry and anxious driving does not have a substantial hazard mitigation performance. It can be explained by the fact that cautious drivers are more accustomed to observing road conditions and making careful driving decisions. Due to this, cautious drivers are more aware of other drivers' conditions than those who focus on their own. They drive carefully out of understanding and concern for the behavior of others rather than being concerned with the risks they face and the secondary rewards of sensation, excitement, etc. They can be said to value caution and apply ethical standards to their driving (Taubman-Ben-Ari & Yehiel, 2012). As a result, they had the highest hazard perception. They performed significantly better than the other three groups in dealing with dangerous situations, as evidenced by the fact that RL and PIVE were shorter than all the other three groups. It is explained by the fact that they performed better in detection, decision-making, and action implementation. The reason is that cautious drivers have a higher attention span for road conditions and a higher sensitivity to them, enabling them to detect changes in road conditions in time.

Dangerous drivers usually show boldness, sensation-seeking, and confidence and are less concerned with other things (Taubman-Ben-Ari & Yehiel, 2012). Some dangerous drivers make dangerous decisions because they are trying to cope with or avoid negative emotions and stress, and in most cases, the decisions are wrong. In this paper, it can be shown by H5 that under time pressure, a lower proportion of dangerous drivers engage in dangerous mitigation behavior or even do not engage in mitigation behavior, which results in a crash.

Angry drivers usually display irritating, hostile attitudes, and behaviors, with a tendency toward aggressive behavior, such as swearing at other drivers, being typical. As a result, angry drivers are also often perceived as less competent, and their behavior is most violent, accompanied by negative emotions such as confrontation, anger, and irritability. While the behavior indicated by this style may be like a dangerous driving style, the difference lies in the underlying motivation. Its drivers are intent on satisfying the need to win over others and vent frustration rather than seeking excitement (Taubman-Ben-Ari et al., 2004). Angry drivers, therefore, choose to drive similarly to dangerous drivers when faced with dangers but are more attentive to road conditions than dangerous drivers and are thus able to detect danger. However, they may be less likely to engage in mitigation behavior due to their own emotions. It may be because drivers in the anger group usually make impulsive decisions when they perceive danger, and most are wrong.

Anxious drivers are often unsure of their driving ability, see other drivers as an additional source of stress, and experience feelings of alertness and nervousness (Taubman-Ben-Ari et al., 2004). When making decisions, there is usually indecision and an inability to make effective decisions. This result can be seen by the longer PIEV time. In general, anxious drivers can perceive danger when faced with a hazardous situation but make decisions, usually with fear and self-denial, showing indecision, and thus are less likely to adopt hazard mitigation behaviors.

Therefore, to improve driving skills in response to hazard perception, drivers may need some training to address this. It is discussed in detail in the next section.

4.2.1. Visual alerts

Earlier studies have shown that training related to visual scanning can help improve safe driving behavior (Hajiseyedjavadi et al., 2018). The presented study investigates visual alerts' effects on drivers' hazard perception abilities with different driving styles. These cues will likely improve drivers' hazard perception and mitigation abilities. The visual alerts in this study were cued four seconds before the hazard's onset.

First, the effect of visual alerts on hazard perception is considered. The results show visual alerts can improve hazard perception in potentially threatening situations. Drivers who received a visual alert cue could perceive the hazard better than drivers in cases without a warning message.

Second, the effect of visual alerts on hazard mitigation is being considered. Analysis of speed showed that the warning group traveled slower than the control group after the warning appeared for the same hazard scenario. This result suggests that visual alerts are effective in helping drivers mitigate hazards, i.e., slow down. However, the visual alerts' effectiveness in the face of a vehicle hazard event was lower, possibly because drivers understood pedestrian warning cues better than vehicle warning cues.

Third, the effect of visual alerts on driving style was considered. This study showed that drivers with visual alerts had a more significant increase in hazard perception when faced with the same hazard scenario as drivers in all three groups. The main manifestations of this were a reduction in speed and an increase in hazard mitigation behavior. Therefore, the results demonstrate that visual alerts improve driver hazard perception, especially for hazard-averse drivers. Therefore, visual alert training can be provided to dangerous drivers to reduce the probability of dangerous events.

4.2.2. Limitations and future directions

In general, this study has limitations in three areas. Firstly, given the better reflection of real-world conditions in VR scenarios, this study faces the same problem as all other VR-based behavioral studies. The ecological validity of the results is challenged due to the possible differences between the virtual environment and reality. Ecological validity refers to the extent to which the perceptions and responses of experimental subjects can be generalized to real-life settings (Cao et al., 2019). It is highly dependent on the sense of the presence of IVE participants, which is determined by the level of realism of IVE (Zou et al., 2017). Future research is needed to improve the IVE, for example, by improving the quality of model rendering and increasing the sense of driver experience.

Secondly, the visual alert time chosen in this paper is 4 s, effective in resolving hazardous events but weak in determining them in the face of vehicle hazards. Therefore, in the future, consideration can be given to shortening the visual alert time and making it less than 4 s.

Thirdly, five hazard scenarios were selected for the study, and there are other scenarios in the driving process. Therefore, other scenarios could be added in the future.

Finally, this study used a low-resolution VR system and will be studied with higher-resolution equipment in the future

5. Conclusion

This study applied VR technology to study the hazard perception of drivers with different driving styles and to improve their hazard perception through visual alert training. The findings show that (1) hazard perception is weakest in drivers who drive dangerously and is most important in drivers who drive cautiously, which does not differ significantly between drivers who drive anxiously or angrily. (2) Visual alerts can help drivers improve their hazard perception and effectively mitigate hazards. (3). Visual alerts improved hazard perception the most for dangerous drivers and were influential in mitigating hazards. These results may have many implications for road safety. For example, our study isolates the variability in hazard perception among drivers with different driving styles. It uses this variability to tailor training to improve hazard perception and reduce the occurrence of hazardous situations. Importantly, we provide visual alert training to improve drivers' hazard perception and reduce the probability of accidents, especially for risky drivers.

CRediT authorship contribution statement

Yan Mao: Conceptualization, Funding acquisition, Methodology, Supervision, Writing – review & editing. Xuan Wang: Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. Wu He: Conceptualization, Methodology, Software, Visualization, Writing – review & editing. Gaofeng Pan: Conceptualization, Software, Visualization, Writing – review & editing.

Data availability

Data will be made available on request.

Acknowledgement

This work was supported by National Natural Science Foundation of China under Grant 72204180. No competition of interests between the various authors directly.

References

Abdu, R., Shinar, D., & Meiran, N. (2012). Situational (state) anger and driving. Transportation Research. Part F, Traffic Psychology and Behaviour, 15, 575–580.
Abele, ¯ L., Haustein, S., Martinussen, L. M., & Mller, M. (2019). Improving drivers' hazard perception in pedestrian-related situations based on a short simulator-based intervention. Transportation Research. Part F, Traffic Psychology and Behaviour, 62, 1–10.
Agrawal, R., Knödler, M., Fisher, D. L., & Samuel, S. (2018). Virtual reality headset training: Can it be used to improve young drivers' latent hazard anticipation and mitigation skills. Transportation Research Record: Journal of the Transportation Research Board, 2672, 20–30.
Ali, Y., & Haque, M. M. (2023). Modeling the response times of mobile phone distracted young drivers: A hybrid approach of decision tree and random parameters duration model. Analytic Methods in Accident Research, 39, Article 100279.
Borowsky, A., Shinar, D., & Oron-Gilad, T. (2010). Age, skill, and hazard perception in driving. Accident Analysis and Prevention, 42(4), 1240–1249.
Boufous, S., Ivers, R., Senserrick, T., & Stevenson, M. (2011). Attempts at the practical on-road driving test and the hazard perception test and the risk of traffic crashes in young drivers. Traffic Injury Prevention, 12(5), 475–482.
Cao, L., Lin, J., & Li, N. (2019). A virtual reality based study of indoor fire evacuation after active or passive spatial exploration. Computers in Human Behavior, 90, 37–45.
Cao, S., Samuel, S., Murzello, Y., Ding, W., Zhang, X., & Niu, J. (2022). Hazard perception in driving: A systematic literature review. Transportation Research Record, 2676(12), 666–690.
Castanier, C., Deroche, T., & Woodman, T. (2013). Theory of planned behaviour and road violations: The moderating influence of perceived behavioural control. Transportation Research. Part F, Traffic Psychology and Behaviour, 18(5), 148–158.
Chen, Q., Chen, Y., Bostrom, O., Ma, Y. H., & Liu, E. Y. (2014). A comparison study of car-to- pedestrian and car-to-e-bike accidents: Data source: The China in-depth accident study (CIDAS). SAE Technical Paper, 2014-01-0519.
Chen, W., Song, J., Wang, Y., Wu, C., Ma, S., Wang, D., & Li, H. (2023). Inattentional blindness to unexpected hazard in augmented reality head-up display assisted driving: The impact of the relative position between. Traffic Injury Prevention, 24(4), 344–351.
Chen, Y., Bostrom, O., Matsushita, T., Ma, L., & Liu, E. Y. (2013). Car to electric two-wheelers and moped accidents analysis: Chinese in-depth real-life data. In Proceeding of 2013 JSAE annual congress (pp. 314–20135062).
Crundall, D. (2016). Hazard prediction discriminates between novice and experienced drivers. Accident Analysis and Prevention, 86, 47–58.
Crundall, D., Loon, E. V., Baguley, T. S., & Kroll, V. (2021). A novel driving assessment combining hazard perception, hazard prediction and theory questions. Accident, Analysis and Prevention, 149.
Elander, J., West, R., & French, D. (1993). Behavioral correlates of individual differences in road-traffic crash risk: An examination method and findings. Psychological Bulletin, 113(2), 279–294.
Fan, X., Pan, G., & Mao, Y. (2019). Investigating the effect of personality on left-turn behaviors in various scenarios to understand the dynamics of driving styles. Traffic Injury Prevention, 20(8), 801–806.
García, J. L. P., Castro, C., Doncel, P., & Taubman-Ben-Ari, O. (2020). Adaptation of the multidimensional driving styles inventory for Spanish drivers: Convergent and predictive validity evidence for detecting safe and unsafe driving styles. Accident Analysis and Prevention, 136, Article 105413.
Hajiseyedjavadi, F., Zhang, T. A., Agrawal, R., et al. (2018). Effectiveness of visual alerts on young drivers hazard anticipation and hazard mitigation abilities. Accident Analysis and Prevention, 116, 41–52.
Haque, M. M., Ohlhauser, A. D., Washington, S., & Boyle, L. N. (2016). Decisions and actions of distracted drivers at the onset of yellow lights. Accident Analysis and Prevention, 96, 290–299.
Horswill, M. S. (2017). Hazard perception tests. Handbook of teen and novice drivers: Research, practice, policy, and directions (pp. 439–450).
Horswill, M. S., & Mckenna, F. P. (2004). Drivers' hazard perception ability: Situation awareness on the road. A cognitive approach to situation awareness. Theory and Application, 9, 155–175.
Horswill, M. S., Hill, A., & Wetton, M. (2015). Can a video-based hazard perception test used for driver licensing predict crash involvement? Accident Analysis and Prevention, 82, 213–219.
Huestegge, L., Skottke, E. M., Anders, S., Müsseler, J., & Debus, G. (2010). The development of hazard perception: Dissociation of visual orientation and hazard processing. Transportation Research. Part F, Traffic Psychology and Behaviour, 13(1), 1–8.
Huo, D., Ma, J., & Chang, R. (2020). Lane-changing-decision characteristics and the allocation of visual attention of drivers with an angry driving style. Transportation Research. Part F, Traffic Psychology and Behaviour, 71, 62–75.
Ishibashi, M. (2007). Indices for characterizing driving style and their relevance to car following behavior. In Sice, conference. IEEE.
Jackson, L., Chapman, P., & Crundall, D. (2009). What happens next? Predicting other road users' behaviour as a function of driving experience and processing time. Ergonomics, 52(2), 154–164.
Kahana-Levy, N., Shavitzky-Golkin, S., Borowsky, A., & Vakil, E. (2019). The effects of repetitive presentation of specific hazards on eye movements in hazard perception training of experienced and young-inexperienced drivers. Accident Analysis and Prevention, 122, 255–267.
Kinateder, M., Ronchi, E., Nilsson, D., Kobes, M., & Muhlberger, A. (2014). Virtual reality for fire evacuation research. In Computer science & information systems. IEEE. Kinateder, M., Comunale, B., & Warren, W. H. (2018). Exit choice in an emergency evacuation scenario is influenced by exit familiarity and neighbor behavior. Safety Science, 106, 170–175.
Li, X., & Sun, J.-Q. (2017). Studies of vehicle lane-changing dynamics and its effect on traffic efficiency, safety and environmental impact. Physica A, 467, 41–58.
Lin, J., Cao, L., & Li, N. (2020). How the completeness of spatial knowledge influences the evacuation behavior of passengers in metro stations: A VR-based experimental study. Automation in Construction, 113, Article 103136.
Long, S., & Ruosong, C. (2019). Reliability and validity of the multidimensional driving style inventory in Chinese drivers. Traffic Injury Prevention.
Lu, G., Wang, Y., Wu, X., & Liu, H. X. (2015). Analysis of yellow-light running at signalized intersections using high-resolution traffic data. Transportation Research. Part A, 73, 39–52.
Malone, S., & Bruenken, R. (2016). The role of ecological validity in hazard perception assessment. Transportation Research. Part F, Traffic Psychology and Behaviour, 40, 91–103.
Mangalore, G. P., Ebadi, Y., Samuel, S., Knodler, M. A., & Fisher, D. L. (2019). The promise of virtual reality headsets: Can they be used to measure accurately drivers hazard anticipation performance? Transportation Research Record, 2673(10), 455–464.
Mckenna, F. P., & Crick, J. (1994). Hazard perception in drivers: A methodology for testing and training. Trl Contractor Report.
Mcknight, A. J., & Mcknight, A. S. (2003). Young novice drivers: Careless or clueless? Accident Analysis and Prevention, 35(6), 921–925.
Mesken, J., Hagenzieker, M. P., Rothengatter, T., & de Waard, D. (2007). Frequency, determinants, and consequences of different drivers' emotions: An on-the-road study using self-reports, (observed) behaviour, and physiology. Transportation Research. Part F, Traffic Psychology and Behaviour, 10, 458–475.

Navon, M., & Taubman-Ben-Ari, O. (2019). Driven by emotions: The association between emotion regulation, forgivingness, and driving styles. Transportation Research. Part F, Traffic Psychology and Behaviour, 65, 1–9.

O'Brien, F., Bible, J., Liu, D., & Simons-Morton, B. G. (2017). Do young drivers become safer after being involved in a collision? Psychological Science, 28, 407–413.

Renge, K. (1998). Drivers' hazard and risk perception, confidence in safe driving, and choice of speed. IATSS Research, 22(2), 103–110.

Ryu, D., Cooke, A. M., Bellomo, E., & Woodman, T. (2020). Watch out for the hazard! Blurring peripheral vision facilitates hazard perception in driving. Accident Analysis and Prevention, 146(105755), 1–13.

Schultheis, M. T., & Mourant, R. R. (2001). Virtual reality and driving: The road to better assessment for cognitively impaired populations. Presence: Teleoperators & Virtual Environments, 10, 431–439.

Scialfa, C. T., Borkenhagen, D., Lyon, J., Deschenes, M., Horswill, M., & Wetton, M. (2012). The effects of driving experience on responses to a static hazard perception test. Accident Analysis and Prevention, 45(1), 547–553.

Sheet, F. (2007). Hazard perception. Fact Sheet.

Slater, M., & Wilbur, S. (1997). A framework for immersive virtual environments (FIVE): Speculations on the role of presence in virtual environments. Presence: Teleoperators & Virtual Environments, 6, 603.

Steuer, J. (2010). Defining virtual reality: Dimensions determining telepresence. Journal of Communications, 42(4).

Sun, L., Cheng, L., & Zhang, Q. (2021). The differences in hazard response time and driving styles of violation-involved and violation-free taxi drivers. Transportation Research. Part F, Traffic Psychology and Behaviour, 82, 178–186.

Taubman-Ben-Ari, O., & Skvirsky, V. (2016). The multidimensional driving style inventory a decade later: Review of the literature and re-evaluation of the scale. Accident Analysis and Prevention.

Taubman-Ben-Ari, O., & Yehiel, D. (2012). Driving styles and their associations with personality and motivation. Accident Analysis and Prevention, 45, 416–422.

Taubman-Ben-Ari, O., Mikulincer, M., & Gillath, O. (2004). The multidimensional driving style inventory–scale construct and validation. Accident Analysis and Prevention, 36(3), 323–332.

Upahita, D. P., Wong, Y. D., & Lum, K. M. (2018). Effect of driving experience and driving inactivity on young driver's hazard mitigation skills. Transportation Research. Part F, Traffic Psychology and Behaviour, 59, 286–297.

Van Huysduynen, H. H., Terken, J., & Eggen, B. (2018). The relationship between self-reported driving style and driving behavior a simulator study. Transportation Research. Part F, Traffic Psychology and Behaviour, 56, 245–255.

Ventsislavova, P., Crundall, D., Baguley, T., Castro, C., Gugliotta, A., Garcia-Fernandez, P., et al. (2019). A comparison of hazard perception and hazard prediction tests across China, Spain, and the UK. Accident Analysis and Prevention, 122, 268–286.

Vlakveld, W., Romoser, M. R. E., et al. (2018). Do crashes and near crashes in simulator-based training enhance novice drivers' visual search for latent hazards? Transportation Research Record, 2265(1), 153–160.

Wells, P., Tong, S., Sexton, B., Grayson, G., & Jones, E. (2008). Cohort II: A study of learner and new drivers. Questionnaires and data tables: Vol. 2. WHO (2018). Global status report on road safety 2018. World Health Organization.

Wilde, G. J. S. (2010). The theory of risk homeostasis: Implications for safety and health. Risk Analysis, 2(4), 209–225.

Ya, A., Zz, A., Mmh, B., My, A., & Sw, A. (2020). Detecting, analysing, and modelling failed lane-changing attempts in traditional and connected environments. Analytic Methods in Accident Research, 28.

Zheng, Z., Ahn, S., Chen, D., & Laval, J. (2011a). Applications of wavelet transform for analysis of freeway traffic: Bottlenecks, transient traffic, and traffic oscillations. Transportation Research. Part B, 45(2), 372–384.

Zheng, Z., Ahn, S., Chen, D., & Laval, J. (2011b). Freeway traffic oscillations: Microscopic analysis of formations and propagations using wavelet transform. Procedia – Social and Behavioral Sciences, 17, 702–716.

Zhou, J., Hirano, M., & Yamakawa, Y. (2022). High-speed recognition of pedestrians out of blind spot with pre-detection of potentially dangerous regions. In 2022 IEEE 25th international conference on intelligent transportation systems (ITSC) (pp. 945–950).

Zou, H., Li, N., & Cao, L. (2017). Emotional response–based approach for assessing the sense of presence of subjects in virtual building evacuation studies. Journal of Computing in Civil Engineering, 31(5).

maoimprovinghazardperceptiondrivers2023.md 91 KB 文件历史 原始文件