Keywords

1 Introduction

Emergency medical service (EMS) for a vehicle accident is a type of emergency service which is designed for responding automobile injuries that require out-of-hospital acute medical care. The goal of the EMS for the vehicle accident is to provide urgent medical care and to move patients to an emergency department at a hospital as fast as possible. The EMS response times often predict the likelihood of a patient’s survival. According to the previous study [1], a minute increase in response time can result in up to a 17% increase in mortality 90 days after an incident involving EMS response. However, it is hard to notice the need for EMS until someone made an emergency call. Also, it is difficult to predict patient conditions until medical crews arrive at the scene. Hence, in this study, we investigated the new way to predict the severity of vehicle accident by analyzing the pattern of pupil dilations. So that, the need for EMS will be automatically sent to 911 operators immediately.

Many researchers have proposed various methods to reduce the EMS response time for car accidents. One study has focused on using geographic information system (GIS) capabilities to determine the optimal placement of emergency response vehicles based on differences in accident density across a city [2]. Another study developed a smartphone app that records collision data and provides approximate values of the external forces caused by accident. The app could then send this information to EMS operators. Also, new technologies, such as OnStar, have already been created to provide an automatic rapid crash response. OnStar, created by General Motors, uses built-in sensors on the car that provides data such as the change in velocity and the directions of the force of the crash. After a collision is detected, an OnStar advisor will contact a driver and emergency responders [3]. However, these technologies rely on data from the vehicle itself. Also, the sensors in a car could be destroyed during the accident. For that reason, it is necessary to develop the method for the driver-centered monitoring rather than the vehicle-centered monitoring to reduce the EMS response time for vehicle accidents.

In this study, we investigated the relationship between driver’s pupil dilations and severity of vehicle crash. According to the previous study [4], the diameter of the pupil changes not only reflexively, as a function of the intensity of light reaching the retina, but also in response to other cognitive functions, such as arousal level, emotion, and attention. Therefore, we hypothesized the changes in pupil diameter could indicate the occurrence of a collision and predict the severity of an accident. In our experiment, we examined pupil diameter changes during vehicle collisions to determine how pupil diameter changes in response to vehicle collisions in a driving simulator environment. We concluded that pupil diameter would increase before rear-ended collisions, and the rate of pupil diameter change was significantly greater for severe crashes compared to minor crashes.

2 Method

2.1 Apparatus

In this study, we used OpenDS (see Fig. 1), which is a cross-platform, open-source driving simulation software [5]. The Logitech G27 driving device (see Fig. 2a) was used as a steering wheel, a break, and an accelerator [6]. OpenDS consists of two major components: the simulator and the drive analyzer. The simulation component can load the driving task which is available in XML formats. The simulation provides multiple modules to redesign scenarios and create complex driving environments.

Fig. 1.
figure 1

Screenshot of OpenDS driving simulation

Full size image
Fig. 2.
figure 2

(a) Logitech G27electronic steering wheel and pedal and (b) Tobii Pro Glasses 2 eye tracking device

Full size image

An eye tracker, which is a device for measuring eye positions and eye movement, was used to examine the data of the driver’s eye movements. This study used the Tobii Pro Glasses 2 eye-tracking device (see Fig. 2b). It is a wearable eye-tracking device which is designed to capture the natural viewing behaviors in a real-world environment. There are three cameras in this device (frame rate 100 Hz). One is a full HD wide-angle scene camera to capture all the details of the surrounding environment. Other two cameras record the gaze movement of both eyes. Tobii Pro Glasses 2 has been used for a wide range of applications, such as measuring cognitive load, performance assessment, and usability studies.

2.2 Participants

A total of twenty college students (Age: 20–22, 18 males and 2 females) participated in this experiment. Their average driving experience was 4 years.

2.3 Experimental Setup

Vehicle accidents data was collected through the driving simulator experiment with two scenarios of different weather conditions. The traffic condition of each scenario was designed based on the previous research [7]. Each scenario was designed to include the event that had the potential cause of car accidents. This event occurred when participants drove onto the ramp and tried to merge onto the highway. As the participants had attempted to merge, the car in front of the participants was decelerated immediately. The speed of the leading vehicle abruptly decreased from 43 miles per hour to 3 miles per hour (see Fig. 3 from position A to position B). Then, the leading vehicle accelerated to merge onto the highway along the dashed line with an arrow. As merging required short-term visual searching, this event significantly increases the risk of vehicle crash. Specifically, when the participants drove into the orange areas shown in Fig. 3, the participants must utilize their divided attention, which is an ability to process two or more responses or react to two or more different demands simultaneously. In this situation, the driver must check a side view mirror to prepare for merging and reacting to the deceleration of the front vehicle. The events were randomly occurred to prevent any driving behaviors caused by effects on the learning of those events.

Fig. 3.
figure 3

The event occurs when participants drive onto the ramp and try to merge onto the highway

Full size image

According to Wu, Abdel-Aty [8], the crash risk was prone to increase at ramp vicinities in fog weather conditions. Therefore, we developed two scenarios for this study: (1) sunny weather and (2) foggy weather. In both scenarios, the participants drove along a local road and drove to the entrance to the highway. After that, they merged onto the highway and drove along the highway itself. Finally, they exited the highway and returned to the local road (see Fig. 4). The participants drove each scenario for 30 min. After that, NASA-TLX questionnaires were asked to the participant in order to measure their workload during the driving simulation. The experiment was conducted in the laboratory (see Fig. 5), where indoor lighting would be turned off once the simulation began. The illumination during the experiment was 24 lx for the sunny scenario and 38 lx for the foggy scenario.

Fig. 4.
figure 4

Route of driving scenario

Full size image
Fig. 5.
figure 5

Setup of experiment

Full size image

2.4 Procedure

Before the experiment, the participants completed a questionnaire to collect demographic information about each participant such as age, gender, driving experience (in years), level of video game experience from novice to expert, and whether the participant had ever been in a car accident and if yes, the severity on a scale of 1 to 10. When finished with the questionnaire, the task for the experiment was given. To further explain and clarify the task, each participant completed a training scenario using a training simulation on the OpenDS simulator. Then, the participants put on the eye tracking glasses. The glasses were calibrated while the participants looked at the center of the calibration card while the card was held up in the center of the middle monitor at a distance of 2.3 feet ahead. The participants were given two scenarios with different weather conditions: a scenario with sunny weather and a scenario with foggy weather. They drove each scenario for 30 min using the driving simulator. Ten minutes of break were given between each scenario. After completing each scenario, they were asked to fill out the NASA-TLX form. After the second scenario and the NASA-TLX survey were completed, the participants were free to leave. The whole experiment took about 90 min (see Fig. 6).

Fig. 6.
figure 6

Procedure of driving simulator experiment

Full size image

3 Results

After the experiment, twenty-six accident data points were collected. They were categorized into two levels based on the severity level of the accident (minor and major level). The minor level indicates a minor car accident, which is a near crash. The major level represents a severe crash, which means that the cars collide and at least one of vehicles is knocked entirely off of the road and possibly flipped over. Figures 7(a) and (b) show the scatterplots of participants’ pupil diameter changes in these two cases (minor and major) data.

Fig. 7.
figure 7

A slope of pupil diameter change reflects the severity of vehicle accidents. Time courses of pupil diameter modulation around the occurrence of car accidents in minor and major severity. The vertical grey line indicates the beginning of the collision moment. Vertical dash lines indicate the driver’s first fixation point of the accident vehicle before the crash accident. Shaded areas, intervals used for estimation of response amplitude. Blue represents minor accidents; orange represents major accidents. Black lines are the linear model lines. (Color figure online)

Full size image

According to the result of the pupil diameter scatterplots, we found that the pupil began to dilate when the subjects indicated that the car in front reduced speed rapidly. The size of the pupil diameter reached to the largest point at the moment of the collision. In addition, the slopes of the linear lines in Fig. 7 were significantly different between the minor level and the major level. The slope of the pupil trace was measured by fitting a line (using linear regression analysis). The trace of pupil diameter size change was highlighted as a function of time. The line started with a driver’s first fixation point of the accident vehicle. The moment of the collision was the end point of the line (see the solid lines in Fig. 7). The results showed that the slope was steeper for more severe accidents. According to the slope comparison between the minor level and the major level, there was a significant difference in pupil size slope for both eyes [left: F (1, 24) = 6.85, P = 0.015; right: F (1, 24) = 6.35, P = 0.019].

Figure 8 shows the result of interval plot of pupil size slope between minor and major vehicle accidents. The result indicated that the mean score of pupil size slope of the major accident (left eye: M = 0.814, SD = 0.305; right eye: M = 0.873, SD = 0.414) was significantly higher than the minor accident (left eye: M = 0.497, SD = 0.247; right eye: M = 0.523, SD = 0.258).

Fig. 8.
figure 8

Interval plot of slope vs accident level

Full size image

4 Discussion and Conclusion

This study compared the driver’s pupil diameter in car accidents with different severity levels in a driving simulation environment. The objective of this research is to understand the relationship between changes in pupil diameter and a vehicle accident. The results indicate the rapid increment in pupil diameter is associated with a more severe vehicle accident. In the experiment, the participants recognized the collision events by noticing the brake lights and the shortened distance to the car in front. Therefore, we used the participants’ first fixation point on the illuminated brake lights of the leading vehicle as the start point for the data analysis. When the participants noticed the brake lights (aware of the possible collision), the pupil diameter started to increase and it stopped at the moment of the vehicle collision. According to the results of the pupil diameter trace, we found the different patterns of pupil dilation caused by different severity levels of vehicle crash. There was a significant difference in the slope of pupil diameter between the major and minor car collisions. When the participants experienced severe accidents, their increment speed of pupil diameter was much faster compared to the minor accidents. Since the hazard events were designed as unexpected decelerating of a leading vehicle (the details in Sect. 2.3), all accident events were all rear-ended accidents. The rear-end collision is a traffic accident wherein vehicle crashes into a leading vehicle. In contrast to other traffic accidents, such as being hit by the following car, the driver can witness the whole process of the vehicle accident. Based on the scene in front of the driver, drivers will have mental accounts for the prediction of the severity level of car accidents.

According to the previous studies, the pupil response is a direct reflection of neurological activities and is associated with specific areas of the brain and nervous system [9]. Fong [10] found that the stimulation of the autonomic nervous system, known for triggering “fight or flight” responses, induces pupil dilation. The fight-or-flight response is defined as a physiological reaction that occurs in response to a perceived harmful event, attack, or threat to survival. Hence, the participants’ pupils could be dilated as the consequence of the “fight or flight” response when the participants were aware of the dangers of driving during the experiment. When the driver first fixated on the car in front, the pupil initially dilated. Then, as the drivers recognized the car in front as a threat due to the imminent collision, the pupil rapidly dilated. In the major accidents, the increment of the pupil size was faster compared to the minor accidents, because the driver recognized the danger of a collision earlier. With the faster response time, the “fight or flight” response started sooner, causing the faster pupil dilation. Therefore, the fast speed of the pupil size increment indicates more severe crashes and result in shorter available response time.

In conclusion, the result showed that the pupil diameter pattern could indicate the level of accident severity. The slope of pupil diameter change in the major vehicle accidents is steeper than in the minor vehicle accidents. The finding of this study provides a new direction of the human-centered monitoring system for automatic crash response and improves the speed of the roadside rescue. This study will contribute to developing an algorithm to detect the accident severity and inform Emergency Medical Personnel to respond car accidents quickly and efficiently.

There are several limitations of the current study. Since the experiment was conducted in a driving simulator environment, we have no real crash data to measure the injury to drivers due to the collisions. Secondly, we did not consider other factors (e.g., fatigue, heavy traffic conditions, age, and others) that might influence the severity of vehicle accidents.

5 Future Work

For the future work, we will increase the number of the participants to collect more severe accident data. Since the severe accidents have the lowest survival rate, it is crucial to collect various car accident data. Also, the different levels of driving experience might contribute different patterns of the pupil diameter [11]. Hence, it is essential to investigate the relationship between the changes in pupil diameter and driving experience. Furthermore, our study only investigated the changes in pupil diameter for the rear-end collision. Other types of accidents might result in different patterns of pupil dilation. So that, it would be necessary to develop scenarios that simulate different types of road accidents. Finally, other factors that affect pupil dilation, such as changes in light intensity, workload, and fatigue, should be considered in a future study. These factors must be incorporated into any prediction of accident severity using pupil diameter.