Drivers’ EEG Responses to Different Distraction Tasks

Driver distraction has been deemed a major cause of traffic accidents. However, drivers’ brain response activities to different distraction types have not been well investigated. The purpose of this study is to investigate the response of electroencephalography (EEG) activities to different distraction tasks. In the conducted simulation tests, three secondary tasks (i.e., a clock task, a 2-back task, and a navigation task) are designed to induce different types of driver distractions. Twenty-four participants are recruited for the designed tests, and differences in drivers’ brain response activities concerning distraction types are investigated. The results show that the differences in comprehensive distraction are more significant than that in single cognitive distraction. Friedman test and post hoc two-tailed Nemenyi test are conducted to further identify the differences in band activities among brain regions. The results show that the theta energy in the frontal lobe is significantly higher than that in other brain regions in distracted driving, whereas the alpha energy in the temporal lobe significantly decreases compared to other brain regions. These results provide theoretical references for the development of distraction detection systems based on EEG signals.


Introduction
Driver distraction is defined as a momentary or transient redirection of driver's attention away from the task of driving toward a thought, activity, object, or person [1]. This frequently happens when interacting with other passengers for example. Additionally, traveling with children can pose a serious distraction to the driver, effectively drawing attention away from the road and causing a disruption in normal driving patterns [2]. Besides, the use of in-vehicle entertainment systems, real-time information systems, and smartphones also leads to a higher risk of distracted driving crashes. The National Highway Traffic Safety Administration (NHTSA) estimated that 9% of fatal crashes, 15% of injury crashes, and 15% of all police-reported motor vehicle crashes were related to driver distraction in 2019 [3]. Hence, a deeper Academic Editor: Zhongxu Hu * Yechen Qin qinyechen@bit.edu.cn 1 understanding of driver distraction can well support accident studies as well as in-vehicle applications. Usually, driver distraction can be categorized into three types, which are visual distraction (eyes off the roadway), cognitive distraction (mental workload involved with thinking about something apart from driving), and manual distraction (a hand off the steering wheel and manipulating a device) [4]. According to the theory of multiple resources [5,6], when the brain resources required by driver distraction are similar to the required resources of normal driving, the distraction effect would be severe. To reduce the number of distraction-affected crashes, it is necessary to investigate and quantify the specific behaviors and reaction mechanisms of drivers in distracted driving.
When distracted from the main driving task, drivers' performance is degraded (e.g., vehicle speed, acceleration, lane keeping, etc.), which will increase the probability of crash occurrence. Research revealed that engaging in secondary tasks might result in lower-speed perception, more obvious lane departures, and slower braking response times [7,8]. A study by Son and Park [9] used the standard deviation of lane position (SDLP) and steering wheel reversal rate (SRR) as driving performance indicators to assess the effect of distraction on driver behavior. They reported an increase in SDLP in visual distraction and the increase in SRR in both visual and cognitive distraction. Boril et al. [10] studied the influence of cognitive distraction on driver behavior and found that cognitive distraction obviously affected drivers' behavioral performance, resulting in a significant reduction in drivers' steering wheel control ability.
Meanwhile, drivers' visual search activities (e.g., head and eye movement) would also be negatively affected by distraction while driving. Azimian et al. [11] conducted a simulation experiment to investigate the effect of distraction caused by mobile phones on drivers' eye movement while driving at roundabouts. Their results showed that driver's gaze time on the roundabout reduced significantly with obviously increased pupil size in driving with phonecaused distraction. Yao et al. [12] evaluated drivers' visual characteristics (e.g., total eye-off-road time, average saccade duration, proportion of long saccade, and saccade frequency) in five secondary tasks and concluded that there were significant differences in the examined measures across distraction tasks. Similarly, Miyaji et al. [13] also reported that the magnitude and frequency of head movement increased while driving with distraction.
To date, various studies have been conducted to detect distracted driving [14][15][16][17][18]. Zhang et al. [14] used a multichannel convolutional neural network to classify drivers' gaze areas and calculated the thresholds of gaze activity as standards to detect drivers' cognitive distraction. AlShalfan and Zakariah [19] designed a new framework based on the Visual Geometry Group (VGG) to recognize driver faces, arms, and hands from camera images and then comprehensively adopted vehicle driving information to recognize driver distraction. Hoang et al. [20] developed a CNNbased architecture to distinguish distracted driving by judging whether a mobile phone was used or the driver's hands were put on the steering wheel. Kashevnik et al. [21] integrated various detection methods from published papers into a comprehensive detection framework to determine different types of distractions. Zuo et al. [22] proposed a framework based on sliding window multi-scale entropy and bidirectional long-term memory network to explore the distraction information in EEG signals. The extracted EEG features were combined with driver behavior characteristics for distraction detection. Zhang et al. [23] ranked the drivingrelated features according to their importance scores and found that the top-ranking features for distraction detection were mostly related to driving performance, eye, and head movements. Most of these studies use advanced machine learning technologies for distraction detection based on images, which is not within the scope of this study. In this paper, we mainly focus on the influences of distracted driving on drivers' electroencephalography (EEG) responses.
Although the influence of distraction on drivers' driving performances and visual search activities has been well investigated in the literature, how distraction would affect drivers' physiological responses is still not well examined, which has been one of the hot research topics about driver distraction in recent years. Researchers used bioelectric signals to examine drivers' physiological responses including EEG, electrocardiogram (ECG), and galvanic skin response (GSR). Among the various bioelectric signals, EEG has been widely used to study and evaluate the corresponding instant responses to distractions because of its direct and strong interpretation capability to describe human beings' brain activities. Alizadeh and Dehzangi [24] captured the changes in EEG dynamics to measure the impact of secondary tasks on the driver's cognitive state during naturalistic driving. They observed that the most significant changes occurred in channels of the frontal and parietal lobes. Almahasneh et al. [25] applied statistical methods to study the effects of different secondary tasks (math and decision-making problems) on EEG response and driving performance. They found that the right frontal cortex was the most affected area during distracted driving, and proposed that the activation of the right frontal cortex could be taken as a spatial index to indicate whether the driver was in a distracted state or not. Karthaus et al. [26] investigated the influences of secondary acoustic and visual stimuli on drivers in driving simulation tasks. Both the senior and junior groups showed larger P2 (parietal lobe) amplitudes and longer P2 latencies in the acoustic-distraction trials. On the contrary, both groups showed decreased P3b (front-central lobe) amplitudes and longer P3b latencies in the visual-distraction trials.

3
This paper examined the effect of different distraction types on drivers' EEG responses and further determined the significant frequency bands in different brain regions that could be used to describe drivers' distraction types. Hence, the main contributions of this study are threefold.
First, the differences in drivers' EEG activities across various distraction types are investigated from the aspects of different frequency bands of brain waves, which is different from most of the previous distraction studies based on EEG signals that mainly focus on cognitive distraction without investigating the differences between different distraction types.
Second, this paper identifies the sensitive EEG features between the examined distraction types, which can provide evidence for EEG-based distraction detection. Most of the previous studies tended to create a distraction detection system to distinguish distracted and non-distracted driving states based on EEG signals from a certain brain region without exploring the influence of distraction on drivers' EEG responses for the selection of sensitive features. The obtained results are expected to improve this situation.
Third, this paper explores drivers' EEG response differences between cognitive distractions with different cognitive loads, as well as between cognitive distraction and comprehensive distraction. The causes of the findings are analyzed and discussed by comprehensively considering brain science knowledge. Although previous studies have reported distraction influences on drivers' EEG response, the differences between cognitive loads and between different distraction types have not been well investigated.
These contributions can provide useful evidence for researchers and engineers to further explore how the human brain works or to design distraction detection applications based on distraction-sensitive EEG signals.
The rest of the paper is structured as follows: Sect. 2 explains the methodology including the test design and signal processing. Section 3 presents the statistical processing results of the collected EEG data, followed by a detailed discussion in Sect. 4. The conclusions are given in Sect. 5.

Participants
As frequently reported in previous studies [27][28][29][30][31], driver gender and age are two of the most influencing factors of driver behavior. To avoid the influence of gender and age effect on the obtained results, only young male drivers are recruited for this study. In total, twenty-four male participants aged from 20 to 26 years old (mean = 22.9; SD = 1.7) are recruited from the local university. They all have a driver's license for at least one year and have normal or corrected-to-normal vision and normal hearing. Every participant self-reports not taking any drugs or drinking alcoholic beverages that might make them drowsy or distractive. All the participants have read and signed the consent procedure that was approved by the local ethics committee.

Apparatus
A fixed-based driving simulator shown in Fig. 1 is used for experiments in this study. The roads and traffic scenarios are designed by UC-Win/Road Ver.8 Standard (FORUM8, Tokyo, Japan). The refreshing rate of the simulation systems is 20 Hz. A 32-channel EEG electrode cap is worn by the participants for data collection with a sampling rate of 1000 Hz. The electrodes are referenced to both of the earlobes, and their connection impedances are calibrated to be less than 5 kΩ. The locations of the EEG electrodes are determined according to the international 10-20 system shown in Fig. 2.

Primary Task
There is a motor vehicle driving at a speed of 110 km/h in the middle lane of a three-lane highway, serving as a target to be followed by the participant. Traffic flows with 800 vehicles/h are designed to fill the left and right lanes to simulate a real traffic environment. The participants are required to follow the lead vehicle at an appropriate following distance while complying with traffic rules at the same time.

Secondary Tasks
In this test, three kinds of secondary tasks are used to induce driver distraction, including a clock task, a 2-back task, and a navigation task. For each task, the subjects are required to give responses as accurately and rapidly as possible on the premise of driving safety for the primary task. Details of the three secondary tasks are as follows.
(1) Clock task: The participants describe the angle (acute, right, or obtuse) between the hour hand and the minute hand for a specific clock time. They have 25 s to respond.
(2) 2-back task: The participants would hear 12 random elements (numbers or letters) in turn. After hearing the third element, they are required to verbally repeat the first element. Similarly, they need to repeat the tenth element when they heard the last one. This task would be completed in 70 s. (3) Navigation task: The participants are required to search the route to a given address in the navigation software (Baidu map) on a mobile phone after hearing the destination information from the voice broadcast. The mobile phone is installed on the right side of the steering wheel (see Fig. 1). Participants are required to complete the task in 60 s.
The choices of distraction tasks are determined according to the research purposes. Specifically, to compare the differences in different cognitive loads, we need to choose two cognitive distractions with different load levels. The 2-back task and clock task are widely used in previous distraction studies [32,33]. During the experiment, the clock task only cost 3-8 s to complete, while the 2-back task cost about 20-35 s, which is a process that requires drivers' constant distraction with extensive brain cognitive resources [34]. Therefore, these two tasks are chosen to examine drivers' EEG responses to distraction tasks with different workloads. Similarly, to compare the differences between cognitive distraction (i.e., 2-back and clock) and comprehensive distraction, mobile phone navigation task is selected because of its complex involvement of multiple distractions including visual, cognitive, and manual operations [35].

Experimental Procedure
Each subject is given a verbal introduction to the experiment at the beginning. Then, the subjects were put on the electrode cap aided by an experimental assistant. A 15-min practice trial is provided for the subjects to be familiar with the driving simulator. Four formal test trials are then conducted for data collection. The duration of each trial is about 15 min, with a 5-min break between every two trials to avoid the subjects getting drowsy. In the first 5 min of each trial, the subjects need to drive normally without any secondary task. In the following 10 min' driving, they are required to complete thirteen secondary tasks (five navigation tasks, five clock tasks, and three 2-back tasks). The time interval between every two consecutive tasks randomly varies from 10 to 50 s. Upon hearing the voice prompt to start the task, subjects pressed the button mounted on the steering wheel. They pressed the button again after they completed the task. Therefore, the EEG data episodes associated with each secondary task could be segmented from the source data according to the recorded button press time. The tasks in the four trials are randomly arranged to minimize the confounding effect of anticipation.

Preprocessing
The EEG data are recorded with a sampling rate of 1000 Hz and then down-sampled to 100 Hz for simplifying data processing. Then, the data are preprocessed using a low pass filter with a cutoff frequency of 50 Hz to remove power noise and other high-frequency noise [36]. EEG artifacts are removed by independent component analysis (ICA), which is a method to separate non-brain signals from EEG. Various artificial components including eye movement, blinking, heart signal, muscle noise, and line noise are separated and removed [37]. After obtaining clean EEG signals, the EEG data corresponding to the distraction tasks are segmented from the time series. In total, 4896 EEG episodes are collected from the source data. The numbers of episodes for the navigation task, clock task, and 2-back task are 960, 480, and 576, respectively. To compare with the episodes without distraction, 2880 non-distraction EEG episodes are also collected from the normal driving data.

Statistical Analysis
Because the EEG data are not normally distributed, a Friedman nonparametric test is used to identify differences in brain activity due to distracting tasks in different brain regions. Furthermore, differences in band energies among the frontal, parietal, occipital, and temporal lobes are also analyzed by the Friedman test. Values with p < 0.05 are considered statistically significant. SPSS statistics analysis software (IBM SPSS Statistics 26) is used for statistical procedures.

Differences Between Distraction Types
The Friedman test is performed to determine whether there are significant differences among driver distractions on the four EEG frequency bands from different brain regions. The post hoc two-tailed Nemenyi test is performed for every Friedman result that is significant to identify where exactly the differences existed. The corresponding results are presented in Table 1. Figure 3 shows the energy values of theta, alpha, and beta activities in different brain regions.

2.74]) is significantly
In the temporal lobe, theta activity in the navigation task (10.06 [11,42]) is significantly higher than non-distraction (6.20 [5.27], p = 0.031) and clock task (5.87 [7.9], p = 0.044). Alpha activity in the navigation task (4.96 [3.55], p = 0.001) increases significantly compared to non-distraction (4.19 [2.77]). Beta activity in the navigation task (10.07 [11.42]) is found significantly higher than non-distraction (6.65 [7.81], p = 0.022), clock task (7.90 [7.56], p = 0.010), and 2-back task (7.13 [7.75], p = 0.032). Table 2 presents the differences in brain regions among different distractions for the four frequency bands. The results show that, except navigation task, delta energy is significantly different among brain regions in the 2-back task (p = 0.031), clock task (p = 0.044), and non-distraction (p = 0.003). Significant differences are found between the frontal lobe and occipital lobe in the 2-back task (p = 0.022), as well as non-distraction (p = 0.001), while the frontal lobe is significantly different from the temporal lobe (p = 0.044) in the clock task. Theta energy shows significant differences among brain regions in all four driving situations (p ≤ 0.001). In the 2-back task, the frontal lobe is found to be significantly different from the parietal lobe (p = 0.005), occipital lobe (p < 0.001), and temporal lobe (p < 0.001). The same situation happens in the clock task (p ≤ 0.001 for all) and the navigation task (p = 0.015, 0.001, 0.044, respectively). The frontal lobe in non-distraction is significantly different from the occipital lobe (p < 0.001) and temporal lobe (p < 0.001). Alpha energy is found significantly different among brain regions in all four driving situations. The post hoc two-tailed Nemenyi test reveals that alpha energy in the temporal lobe is significantly lower than in the other brain regions (p ≤ 0.001 for all). In non-distraction driving, alpha energy in the frontal lobe is significantly lower than parietal lobe (p = 0.012) and occipital lobe (p = 0.007).

Differences Among Brain Regions While Driving with Different Distractions
For beta energy, except non-distraction, significant differences are found in clock task (p = 0.001), 2-back task (p < 0.001), and navigation task (p = 0.035). Beta energy in the frontal lobe is significantly higher than the other brain regions (p = 0.010, 0.001, and 0.002, respectively) in the clock task, as well as the 2-back task (p = 0.001, 0.002, and 0.001, respectively). As for the navigation task, the occipital lobe is significantly different from the temporal lobe (p = 0.022).

Discussion
In recent years, researchers have taken a great interest in investigating the relationship between EEG and specific human behaviors using the most advanced equipment. This work is to analyze the influence of different distraction types and brain workload on driving from the aspect of EEG response, aiming to consolidate and improve the current knowledge on applying EEG data for driver distraction detection. This section discusses the effect of secondary tasks on EEG activity first and then talks about the problems of individual differences in brain activity. The limitations of this study and the corresponding further work are proposed.

Impact of Secondary Tasks on EEG Activity
Brain dynamics in the process of multi-task simultaneous processing is considered to show abundant activation [40]. These significant changes illustrated the increase of human cognitive load, which reflected the distraction caused by secondary tasks. Drivers needed to consume more brain resources to complete multiple tasks at the same time. The clock task and the 2-back task are designed to induce cognitive distraction and increase brain load. The 2-back task mainly occupied human driver's auditory working memory, while the clock task mainly occupied the visual-spatial working memory [41]. The results show that these two kinds of cognitive distractions affect the activities of brain frequency bands in varying degrees, caused by the increase of cognitive workload by secondary tasks. The results suggest that the 2-back task has a more significant impact on changes in brain activities compared to the clock task. The increase or decrease is more significant and affects more brain regions comparatively, which could be explained by the workload of the task. It is obviously easier to complete the clock task than the 2-back task, leading to relatively less brain workload. In the experiment, the clock task could be completed within 5 s, while the 2-back task is a process of exerting brain workload continuously. The frontal lobe is mainly related to attention, learning, working memory, problem-solving, and decision-making [42]. As shown in Fig. 3a, the theta and beta activity of the frontal lobe increases obviously when performing cognitive tasks. The increase in theta is related to mental workload, problem-solving, coding, or self-monitoring process [43]. This result is in line with another study indicating that the increase of theta bands could be considered an index of distraction [44,45]. It suggested that the dual tasks induced more EEG activity in theta and beta bands. In other words, drivers consumed more brain sources to accomplish dual tasks at the same time. Moreover, the results show that the alpha activity decreases significantly in the parietal and occipital lobe, indicating high attention to specific tasks. It counters the results found in the frontal lobe, where alpha activity increased in the 2-back task. Klimesch [46] also reported that the increase of alpha in the frontal lobe was an index of sustained cognitive workload.
The navigation task in this study is a complex multimodal task including visual searching, cognitive thinking, making decisions, and manual operation [47,48]. It belongs to comprehensive distraction and was more complicated than a single cognitive distraction. According to Wickens's multiple resources theory [49], the distracting interference caused by the stimuli should be intense when the brain resources required by these distracting stimuli are more similar to normal driving. Obviously, the visual, spatial, auditory, and other resources required by the navigation task using cellphones are multi-aspect conflicting with driving. In theory, the changes in EEG caused by the navigation task are more significantly different. The statistical results prove this inference. Table 1 shows that significant differences could be found between the navigation task and non-distraction for theta, alpha, and beta activity in four brain regions. The occipital lobe correlates with visual information processing, so the increase of theta activity in the occipital lobe during the navigation task indicates that it requires more visual processing resources.

Individual Differences in Brain Activities
The state of brain waves may be different across subjects and individual differences may need to be considered for EEGrelated studies. Figure 4 compares the brain wave energy of the frontal lobe across the recruited twenty-four participants. The results show that the total energy of brain waves and the ratio of each frequency band are quite different among individuals. For example, the total energy of participant No.18 is nearly twice Fig. 4 number of participant No.21. Participant No.1 has the highest proportion of theta in total energy, and participant No.3 has the highest proportion of alpha. Therefore, the influence of individual differences should be carefully considered in the future development of distraction detection methodologies and practical applications.

Potential Applications
In the development of autonomous vehicles with L3 intelligence, drivers are serving the role of a monitor of the traffic environment and are required to take over the vehicle when encountering emergency situations [50]. Therefore, effectively monitoring drivers' distraction status and deploying interference strategies to pull drivers back from distraction for timely takeover control is critical for driving safety with L3 intelligence. Meanwhile, with the development of EEG data acquisition systems based on dry-wireless or non-contact sensors, applying the obtained knowledge for distraction detection in actual driving may become feasible.
One of the possible frameworks for future applications is illustrated in Fig. 5. Based on the collected EEG data, the reported sensitive features of driver distraction in this study can be calculated. With the obtained features as inputs, Fig. 4 Energy values of theta, alpha, beta activity in the frontal lobe across the twenty-four participants driver distraction models can be trained based on classical machine learning methods [18,21] or advanced deep learning algorithms [16,17,22,51]. It has been reported in previous studies that drivers' individual differences affect their EEG responses [51]. Therefore, individualized distraction models could be established for better detection performances. According to the detection results, interference strategies (e.g., audio warning) could be triggered to pull drivers back from distraction for safe driving when necessary.

Limitations and Further Work
The limitations and the corresponding future work are discussed from the following perspectives. First, many studies have reported that gender and age affect human EEG activities [27][28][29][30][31]. To exclude the interference of these factors in this study, only young male drivers are recruited to investigate their EEG responses to different distraction tasks, which will limit the generalization of the obtained results. Studies have frequently reported that driver gender and age affect their driving behavior and performance [26,52]. In future work, the influence of gender and age on drivers' EEG responses to different distraction tasks should be further investigated for a more comprehensive understanding. Second, only power spectrum density is examined in this study. Band power [24], band spindles [53], oscillation [40], and amplitude [25] are also useful indicators to evaluate brain activity. In future research, more indicators should be comprehensively analyzed. Third, although the adopted secondary tasks are commonly used in distraction studies, the task may still be somewhat different from real-world distractions in naturalistic driving. Collecting sufficient EEG data by wearable devices while driving with distraction is still a significant challenge for driving safety improvement, which needs further investigation. Fourth, the main driving task in this study is pre-defined and the driving scenario is not very complex. Further investigation on brain activities under near-real complex driving conditions with more flexibility for independent driving should be considered in future studies.
Although several companies are developing dry-wireless electrodes for EEG data collection, which can be worn directly on the head like a common hat, the technology is still not mature enough [54]. Besides, wearing a cap for EEG data collection is not practical in real-world driving. Therefore, applying the obtained results in practical applications based on non-contact EEG acquisition systems is the most ideal solution, but it still faces many technical challenges that need to be solved in future studies.

Conclusions
In this study, three kinds of secondary tasks, including clock task, 2-back task, and navigation task, are conducted in simulator tests to explore brain dynamics in frontal, parietal, occipital, and temporal lobes. The statistical analysis supports the conclusion that comprehensive distraction has a more significant impact on brain activities than single distraction type, which is due to the fact that comprehensive distraction occupies more physical resources. Although the clock task and the 2-back task both belong to cognitive distraction, the brain response difference in the 2-back task is more significant. It indicates that a heavy workload would lead to a stronger distraction effect. In the process of cognitive distraction driving, the theta and beta activities in the frontal lobe increase, whereas the alpha activity in the parietal lobe and occipital lobe decreases. These features can be utilized as effective indicators for detecting driver distraction. The individual differences between subjects indicate that individualized applications should be developed in future intelligent systems.