Portal de Periódicos da CAPES

Determination of optimal electroencephalography recording locations for detecting drowsy driving

2018; Institution of Engineering and Technology; Volume: 12; Issue: 5 Linguagem: Inglês

10.1049/iet-its.2017.0083

1751-9578

Chaofei Zhang, Wenjun Wang, Chaoyang Chen, Chao Zeng, Dennis Anderson, Bo Cheng,

Ergonomics and Musculoskeletal Disorders

IET Intelligent Transport SystemsVolume 12, Issue 5 p. 345-350 Research ArticleFree Access Determination of optimal electroencephalography recording locations for detecting drowsy driving Chaofei Zhang, Chaofei Zhang Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this authorWenjun Wang, Corresponding Author Wenjun Wang wangxiaowenjun@tsinghua.edu.cn Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this authorChaoyang Chen, Chaoyang Chen Department of Biomedical Engineering, Wayne State University, Detroit, MI, 48201 USASearch for more papers by this authorChao Zeng, Chao Zeng College of Information Science and Technology, Shihezi University, Shihezi, 832000 People's Republic of ChinaSearch for more papers by this authorDennis E. Anderson, Dennis E. Anderson Beth Israel Deaconess Medical Center, Boston, USA Harvard Medical School, Boston, USASearch for more papers by this authorBo Cheng, Bo Cheng Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this author Chaofei Zhang, Chaofei Zhang Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this authorWenjun Wang, Corresponding Author Wenjun Wang wangxiaowenjun@tsinghua.edu.cn Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this authorChaoyang Chen, Chaoyang Chen Department of Biomedical Engineering, Wayne State University, Detroit, MI, 48201 USASearch for more papers by this authorChao Zeng, Chao Zeng College of Information Science and Technology, Shihezi University, Shihezi, 832000 People's Republic of ChinaSearch for more papers by this authorDennis E. Anderson, Dennis E. Anderson Beth Israel Deaconess Medical Center, Boston, USA Harvard Medical School, Boston, USASearch for more papers by this authorBo Cheng, Bo Cheng Department of Automotive Engineering, Tsinghua University, Beijing, 100084 People's Republic of ChinaSearch for more papers by this author First published: 06 March 2018 https://doi.org/10.1049/iet-its.2017.0083Citations: 4AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Early detection of drowsy driving is an important issue for driving safety. Quantitative electroencephalography (EEG) is an attractive method for detecting brain activity changes. However, further study is still needed to evaluate the feasibility of wearable devices that can detect drowsy driving in real-world settings. This study sought to determine whether convenient EEG recording locations are sensitive in detecting brain activity changes associated with drowsy driving and to characterise these EEG changes. Twenty-two healthy adult subjects were recruited to participate in a car-following task using a driving simulator. EEG data were recorded from four locations, two frontals (Fp1, Fp2) and two temporals (T3, T4) of the brain while driving. The results showed that the increase of δ activity, decrease of θ and α activity and a decrease of spectral edge frequency at 90% were found in the drowsy state compared to the alert state (paired t -tests, p < 0.05). Effect sizes for EEG changes were larger at the temporal locations compared to frontal locations. This suggests temporal locations can be feasible recording locations for wearable monitoring devices to detect drowsy driving. 1 Introduction Drowsy driving is a constant occupational hazard for drivers. According to the National Sleep Foundation's 2009 annual Sleep in America survey [1], 28% of drivers report driving while drowsy at least once per month in the past year. Ten percent of drivers admitted having fallen asleep while driving in the past year and 41% at some point in their lifetime [2]. Furthermore, studies have concluded that 15–20% of fatal car accidents are fatigue related [3]. Therefore, better detection of driving fatigue is a crucial issue for reducing the number of lives lost due to drowsy driving. Several approaches have been proposed to detect driving fatigue, including facial expression evaluation [4, 5], measures of heart rate variability [6], lane departure monitoring [7] and encephalography (EEG) power spectra [8, 9]. Facial expression evaluation seeks to identify drowsiness from video of drivers' faces, but it is easily affected by light and atmosphere. Lane-departure monitors vehicle operation as an indirect measure of the driver's cognitive state; however, it has difficulty distinguishing between drowsy and non-drowsy states, often accompanied with false-positive warnings. On the other hand, the use of EEG provides a relatively accurate measurement of driver brain activity [5, 8]. Many studies have shown that the brain dynamics linked to fatigue and behavioural lapses can be assessed by EEG power spectra [7, 10-15]. Peiris et al. [7] found that spectral power was higher during drowsiness in the δ, θ and α bands, and lower in the β and γ bands, but correlations between changes in EEG power and lapses were low. Jap et al. [12] found stable δ and θ activities over time, a slight decrease of α activity and a significant decrease of β activity after driving about 60 min. Jap also reported significant differences for α and θ activities at the frontal site and significant differences for δ and θ activities at the temporal site during drowsy driving [11]. Akerstedt et al. [13] found the early stage of drowsiness can be indicated by an increase in θ activity. Stern et al. [14] reported that α activity reflected a relaxed wakeful state, and decreased with concentration, stimulation or visual fixation. However, Akerstedt [13] and Torsvall [15] reported an increase in α activity in train drivers who were sleepy enough to fall asleep while driving. Torsvall [15] suggested that α activity was the most sensitive measure for detecting fatigue, followed by θ and δ activity. Eichele et al. [10] reported EEG and fMRI findings that precede human performance errors and suggested that monitoring of the identified precursor states may help in avoiding human errors in critical real-world situations. Several algorithms and systems have been proposed to predict the drowsy state from EEG data [12, 16-19]. Using an independent component analysis method based on fuzzy neural networks methodology [17] and the comparison of three neural networks [19], Lin [18, 20] demonstrated an automatic drowsiness prediction system with EEG power spectra by constructing a linear regression model. Eoh et al. [9] proposed two ratios of EEG bands ((α + θ)/β and β /α) to be used in a mental fatigue detection technique. Moreover, Jap et al. [12] investigated the performance of four ratios of EEG components ((α + θ)/β, α /β, (α + θ)/(α + β) and θ /β), demonstrating that they may be used for mental fatigue detection. Although several methods [4-6, 8] have been proposed for driver fatigue monitoring, debates on the efficiency of these methods still exist. Commercial products for driver fatigue monitoring are primarily video-based systems. Importantly, there is not yet an EEG-based wearable device to monitor drowsiness. Current EEG parameters may not have sufficient sensitivity and specificity to reliably detect mental fatigue. The aim of this study was to characterise EEG parameters, including power spectrum density (PSD) and spectrum edge frequency at 90% (SEF90), during alert and drowsy driving states, and to determine which EEG recording locations can better detect brain activity associated with drowsiness while driving. 2 Materials and methods Twenty-two healthy college students (16 men and 6 women, mean age of 23.5 ± 3) were recruited to participate in the experiment using a driving simulator (Fig. 1). All participants had normal neurological functioning, normal visual acuity, and a valid driver's license. All procedures were approved by the University Research Ethics Committee. A lycra stretch cap (CAP100C, Biopac Inc., Goleta, CA) with the international 10–20 electrode system was used for EGG recording with a unipolar reference placed at the right earlobe and the ground placed at the central zero (Cz) location of the brain. Ag/AgCl electrodes were used for EEG recording. When the cap was in place, EEG recording gel (Gel100A, Biopac Inc.) was injected into each electrode via a central gel access hole. EEG data were recorded by the BN-EEG2 BioNomadix Wireless EEG transmitter/receiver set, connected to MP-150 data acquisition system (Biopac Inc., Goleta, CA). Four channels of EEG data were recorded from the left frontal (Fp1), right frontal (Fp2), left temporal (T3) and right temporal (T4) locations, respectively, as shown in Figs. 1b and d. The data sampling rate was 1000 Hz. Fig. 1Open in figure viewerPowerPoint Schematic diagram of experiment (a) Driving simulator, (b) EEG recording locations (frontal view), (c) Driving simulating scenario, (d) EEG recording locations (lateral view) Virtual reality-based monotonous highway driving experiments were performed in a driving simulator in Tsinghua University [21, 22] that mimicked realistic driving situations in a temperature-controlled room (Fig. 1a). A passenger car (BMW sedan) was mounted on a six degree-of-freedom motion base, providing a realistic driving experience to subjects. The angular and longitudinal moving range of the vehicle was 0° ± 15° and 0 ± 0.4 m, respectively. The audio simulation unit contains a stereo speaker system to simulate the sound of engine, wind and traffic noises as in real driving. The driving scenario was projected onto five screens: three for front view with a total of 200° field of view and two for rear view with a total of 55° field of view. Vehicle positions and driving performance data (speed, acceleration/deceleration, steering wheel angle etc.) were recorded by a computer [23]. Participants were asked to refrain from consuming caffeine and tea as well as smoking on the testing day and reported compliance with these instructions. The simulated driving tests were performed between 10:00 and 17:00. Moller [24] also used this experimental time of the day in a prior study of driver drowsiness, and its validity had already been confirmed. The driver was asked to follow a front car with a constant speed of 60 km/h on a unidirectional three-lane highway, as shown in Fig. 1c. The subject could stop when they felt too drowsy to keep driving. Before the subjects drove the simulation car, EEG signals were recorded for 5 min, and EEG data were also recorded during the whole experiment period. Two cameras mounted on the dashboard were used to monitor facial expression associated with drowsiness during driving. Three minutes of EEG raw data from the beginning of the driving test (alert state Fig. 2a) were extracted for EEG analysis, as were 3 min of data near the end of the driving test (drowsy state Fig. 2b) which were selected to evaluate EEG changes at the drowsy driving status. Generally, the raw signals of the EEG larger than 50–70 μV are treated as artefact [9]. The gain in this study was 1000, thus measured EEG signals which were larger than 70 mV were manually removed by visually inspecting the EEG raw data. All four channels of the EEG data were then sectioned into 5 s epochs, and were subjected to PSD analysis using Matlab software (The MathWorks, Inc., Natick, MA) to derive the four frequency components of interest, which were δ (0–4 Hz), θ (4–8 Hz), α (8–13 Hz) and β (13–30 Hz) [16]. δ and θ are considered slow wave activities, whereas α and β are considered fast wave activities. SEF90 was also analysed for the EEG data recorded at the four different locations. The PSD analysis generated power spectra magnitude µV2 /Hz. The area under the curve of the spectral magnitude was computed for each frequency band of interest for each of the frequency components (Fig. 3). Fig. 2Open in figure viewerPowerPoint EEG raw data of a representative driver (a) Raw data of alert state, (b) Raw data of drowsy state Fig. 3Open in figure viewerPowerPoint Power spectra density of EEG (PSD), EEG frequency bands, and SEF 90 In this analysis, PSD of EEG was calculated using the Welch method [25] ((1) and Fig. 3). In (1), P (ω) refers to PSD of a section of signal, which is divided into L pieces of length M, and the adjacent pieces are partially overlapping to get better variance performance. is a normalisation factor to make sure the PSD was unbiased estimation. Hanning window was applied. In particular for this study, the PSD was calculated for the 3 min sections of signal representing alert and drowsy states with 2 s windows of length (2000 data points), providing a percentage of overlap was 50% (1) (2) EEG characteristics including δ (0.5–4 Hz), θ (4–8 Hz), α (8–13 Hz), β (13–30 Hz) power spectra and frequency feature analysis in terms of SEF90 (2) were analysed using Matlab software. Then δ, θ, α and β power were normalised by the total power (0.5–30 Hz) to adjust for individual differences in total power (3). In (3), refers to the normalised EEG sub-band power, also described as relative power [26], and represents δ, θ, α and β powers. Moreover, two EEG ratios ((4) and (5)) were also calculated as (3) (4) (5) Thus, the outcome variables examined were relative (normalised) δ, θ, α and β powers, SEF90 and ratios and , determined for two states (alert and drowsy). One-way ANOVA was used to test for differences between alert and drowsy driving states. Each outcome variable from EEG was analysed individually. The significance of driving states at each recording location was calculated using paired t -test. 3 Results Average driving duration on the simulator was 60 ± 10 min. Subjects reported deep drowsiness at the moment they stopped driving. According to observer rating method [27] using recorded facial video of the drivers, the state of drivers at the beginning and end were identified as alert and drowsy, respectively. Relative powers of EEG sub-band activities are summarised in Table 1. Outcomes showed that there were differences between states for the δ band, θ band and SEF90 at various recording locations. Overall, δ activity increased during drowsy state compared to alert state, while θ and SEF90 decreased in the drowsy state. β and α activities appeared to decrease, but there was not a statistical difference at frontal locations. There was not a significant effect on ratio (α + θ)/β, while ratio (δ + θ)/(α + β) decreased at the temporal locations. Table 1. Relative powers of EEG sub-band activity (%), SEF90 and EEG ratios in alert and drowsy driving states. Differences between states by paired t -tests are noted Fp1 Fp2 T3 T4 δ power alert 85.8 ± 3.6 85.5 ± 3.5 75.3 ± 7.1 75.9 ± 8.8 drowsy 89.5 ± 3.4 88.4 ± 3.2 85.0 ± 6.3 83.8 ± 7.0 p -value ** ** ** ** θ power alert 10.7 ± 2.5 10.8 ± 2.2 10.1 ± 2.9 10.0 ± 3.5 drowsy 7.8 ± 2.4 8.4 ± 2.3 6.7 ± 2.7 7.5 ± 3.2 p -value ** ** ** * α power alert 2.0 ± 1.0 2.1 ± 1.0 5.4 ± 1.9 5.0 ± 2.1 drowsy 1.4 ± 0.8 1.6 ± 1.0 3.7 ± 1.7 3.7 ± 2.0 p -value — — * — β power alert 1.5 ± 1.0 1.7 ± 1.2 9.2 ± 5.8 9.0 ± 5.5 drowsy 1.3 ± 1.2 1.5 ± 1.4 4.6 ± 3.1 5.0 ± 3.2 p -value — — * * SEF90 alert 4.7 ± 0.8 4.9 ± 1.0 11.8 ± 5.0 10.9 ± 5.3 drowsy 4.0 ± 0.8 4.3 ± 0.8 7.0 ± 3.6 7.2 ± 4.2 p -value * — ** * alert 14.5 ± 13.6 12.4 ± 9.3 2.8 ± 2.6 2.1 ± 1.1 drowsy 14.2 ± 11.5 12.8 ± 9.5 3.0 ± 1.6 2.8 ± 1.2 p -value — — — — alert 38.0 ± 22.7 35.1 ± 20.6 8.1 ± 5.8 7.8 ± 3.7 drowsy 55.2 ± 34.6 45.7 ± 27.4 15.4 ± 10.5 14.1 ± 7.5 p -value — — * * ** p < 0.01, * p < 0.05, — p > 0.05. Overall, EEG sub-band activities presented various differences between alert and drowsy state at different recording locations. δ and θ activity showed strong effect sizes (paired t -tests, p < 0.01), at the frontal head. δ, θ and SEF90 were showed strong effect sizes at temporal locations (paired t -tests, p < 0.01). α, β, (α + θ)/β and (δ + θ)/(α + β) had weaker or no effects. δ band relative power (%) increased from alert to fatigue state at temporal and frontal-parietal sites (Fig. 4, top). There was a difference between alert and drowsy state at all the front (Fp1 p = 0.01; Fp2 p = 0.008) and temporal locations (T3 p = 0.000; T4 p = 0.009). Fig. 4Open in figure viewerPowerPoint Changes of δ activity (top) and θ activity (bottom) from alert to drowsy state (mean ± SEM) θ band relative power (%) decreased from alert to fatigue state at temporal and frontal-parietal sites (Fig. 4, bottom). There was a difference between alert and drowsy state at all four recording locations (paired t -tests, Fp1 p = 0.003; Fp2 p = 0.002; T3 p = 0.005; T4 p = 0.037). α band relative power seems to decrease during drowsy state compared to alert state (Fig. 5, top). However, there was a significant difference between alert and drowsy state only at T3 (paired t -test, p = 0.018). Fig. 5Open in figure viewerPowerPoint Changes of α activity (top) and β activity (bottom) from alert to drowsy state (mean ± SEM) β band relative power (Fig. 5, bottom) decreases significantly at the temporal locations in the drowsy state (paired t -test, T3 p = 0.015; T4 p = 0.02), but there was no difference at frontal locations (paired t -test, p > 0.05). SEF90 (Fig. 6) decreased significantly from alert to drowsy state at Fp1 (p = 0.02), T3 (p = 0.002) and T4 (0.007) locations, but not in the Fp2 location (p > 0.05). Fig. 6Open in figure viewerPowerPoint Changes of SEF90 from alert to drowsy state There was no difference between alert and drowsy state for the ratio (α + θ)/β at any recording locations (Fig. 7, top, p > 0.05), while the ratio (δ + θ)/(α + β) has a significant increase at the temporal locations (Fig. 7, bottom) (paired t -test, T3 p = 0.02; T4 p = 0.01), but not the frontal locations (p > 0.05). Fig. 7Open in figure viewerPowerPoint Changes of (α + θ)/β(top) and (δ + θ)/(α + β) (bottom) from alert to drowsy state (mean ± SEM) 4 Discussion Using a driving simulator, the current study investigated EEG activity in alert and drowsy driving states and in frontal and temporal EEG recording locations to determine the feasibility of using EEG parameters in detecting drowsiness during driving. In summary, slow wave activities such as δ and θ bands were more sensitive to changes of driver's state compared with fast wave activities including α and β bands. In addition, the proposed feature SEF90 was also a sensitive index to detect changes of EEG during monotonous driving. Changes of δ band relative power at temporal locations (T3 and T4) were more significant than that of frontal head locations (Fp1 and Fp2). Most of these EEG parameters changed significantly with drowsiness at the temporal recording locations, which could reflect a significant change of neuronal activity at the recording location. The temporal lobe is the auditory cortex that is directly connected to the ears and specialises in hearing. In addition, motor cortex and somatosensory cortex are adjacent to the temporal lobe. During alert state, these cortex areas are more significantly involved in cognitive information processing and motor functional performance control than other brain locations. During drowsiness, neuron activity in this cortex is decreased significantly which can lead to corresponding significant changes in EEG. Our study indicated that temporal locations may be a sensitive area for detecting mental changes from the alert to the drowsy state using EEG techniques. A driving simulator was used to conduct the experiment while maintaining driving safety. It can be effectively used to produce sleepiness and drowsy driving [28, 29], but eliminates the risks associated with drowsy driving on real roads. Compared with real-world driving, the subjects faced fewer events and less overall information to be processed. The reduced stimulation in the driving simulator could lead to performance decrements much earlier [30]. In our study, the subjects reported that they felt very drowsy and would fall in sleep if they kept operating the vehicle. The average driving duration on the simulator was 60 ± 10 min. In other studies, it has been reported that 30 min of monotonous driving can induce fatigue [16], and that 90 min driving on a simulator causes mental fatigue with decreased response time in operation [27]. Our results coincide with these research findings, indicative of a successful production of a drowsy driving state. Our study found that δ band activities changed significantly during the drowsy state compared with the alert state. A previous study also reported similar results based on spectral power analysis, demonstrating higher δ power during lapses, an increase of slow wave band activity, and a decrease of fast wave bands activity [7]. In this study, decreased θ power was found in the drowsy state in all four recording locations. This is different from other studies [7, 31, 32] which showed an increase of θ activity. Specifically, Peiris reported increased θ power during drowsiness accompanied by operation lapses [7], while Belyavin and Wright [31] and Lal and Craig [32] reported an increase in both δ and θ during fatigue. Moreover, stable δ and θ activities were reported for long-duration driving on a simulator [4]. These differences could be due to different methods in EEG analysis. Our study used normalised θ band power, showing a decrease of θ activity. If θ band power was not normalised, the average θ power increased but the statistical analysis did not show a statistical difference (paired t -test, p = 0.7). It was found that α band activities significantly decreased only at the T3 location. Measured outcomes of α power vary among existing studies. Akerstedt [13] and Torsvall [15] found an increase in α activity in very sleepy train drivers. Torsvall also suggested that α activity was the most sensitive measure that could be used in detecting fatigue, followed by θ and δ activity. Jap found increased α activity during a long simulator driving [12]. In Eoh's study, α activity showed a decreasing trend, where it had a slightly decreasing pattern, instead of an increasing slope [9]. In this study, β activity in the drowsy state decreased significantly at the temporal locations but not at the frontal locations which had large standard deviations relative to the mean. This indicated that β activity measurement had great variability among subjects. However, the trend was similar to previous studies that reported a decrease of β activity during drowsiness [12, 26]. The results showed a decrease of SEF90 during drowsy state at all the locations. In clinical applications, SEF90 has been used to monitor the depth of anaesthesia, and decreases about 30% from the awake to anaesthetised state [33]. A similar magnitude change between alert and drowsy states was shown in this study for SEF90 at temporal locations. Decreased SEF90 and other EEG bands' depressed activity and relative decrease of sympathetic activity were found in healthy sleeping subjects [34, 35]. Thus, based on our findings and other studies, SEF90 may be a good candidate parameter for detecting drowsy driving. The ratio of slow wave activity to fast wave activity was reported to decrease as driver's drowsiness progressed [20]. Jap [12] found that ratios (α + θ)/β and (δ + θ)/(α + β) showed an increase from alert to drowsy state at all measured locations including temporal (T3, T4) and frontal (Fp1, Fp2) locations. However, in our study, the ratio (α + θ)/β showed no significant difference between alert and drowsy state at all the four recording locations. Furthermore, the ratio (δ + θ)/(α + β) decreased significantly during drowsy state only at T3 and T4 locations. The disagreements may be caused by experiment conditions and individual factors. The experiments of Jap et al. [12] were conducted from 10:30 to 13:30 (near noon, ±1.5 h), and the ages of recruited subjects were 20–70 years (28 ± 10 years), while our experiments were conducted from 10:00 to 17:00, and the subjects were mostly college students aged from 20 to 27 who had less driving experience. On the other hand, the individual factors may be another reason causing the controversial result. Abtahi [30] suggested individual factors as a main reason for controversial results while using ECG to detect driver's sleepiness. As can be seen in Table 1, the variances of (δ + θ)/(α + β), and especially (α + θ)/β, were large, thus there was no significant differences between alert and drowsy states. The effectiveness of different recording locations for drowsiness measured by EEG was explored in this study. Compared with frontal head locations (Fp1 and Fp2), temporal locations (T3 and T4) had more significant features. Moreover, the same features of temporal locations were more significant than that of frontal head locations (Table 1). Thus we suggest temporal locations as the most feasible recording location for wearable devices to monitor drowsy driving. This could also be an explanation for Shen's report [36], which found no key features from Fp1 and Fp2 locations because of the artifices of the electrooculography and electromyography. The difference of our results from other studies might be influenced by external factors such as time on task, light level and driving simulation setting. Due to these factors, it is possible that some effects are being hidden by the length of events, the stimulating effect of getting into the simulator, or discrete events in the driving experience. However, δ and θ power changes are concordantly found in our and other studies, suggesting that they can be reliable indices for detecting drowsy driving. Overall, however, monitoring multiple variables to detect drowsiness may increase the accuracy of detection, which should be explored in future studies. There are additional limitations in this study that should be noted. One is that only four recording locations were selected. There were two main reasons for this: (i) the experimental devices used allowed a limited number of recording locations; (ii) the frontal and temporal locations are likely more convenient for wearable EEG monitoring devices, prioritising the examination of these locations. Another limitation is the generalisability of the findings remains unknown; thus, the conclusion should be tested in real vehicles, real-world driving scenarios and in drivers across a broad range of ages and experience levels. 5 Conclusion Quantitative EEG methods were used to measure drowsy driving during a simulated driving task. The relative power spectra of EEG sub-bands showed statistically significant differences before and after long-term driving. δ activity increased during drowsy state, while θ activity, α activity and SEF90 decreased. Stronger effects were found at temporal locations than frontal locations, indicative that temporal locations may be preferred for wearable devices to monitor drowsy driving. 6 Acknowledgments This research was supported by National Natural Science Foundation of China (grant nos. 51575303, U1664263 and 51565051), and Science and Technology R&D Program of Shihezi University (grant no. RCZX201437). 7 References 1https://sleepfoundation.org/sites/default/files/2009%20SLEEP%20IN%20AMERICA%20SOF%20EMBARGOED_0.PDF, accessed November 2017 2Lindsay S. Tefft B.: 2012 Traffic Safety Culture Index: Motorists Admit to Driving Drowsy, ( AAA Foundation for Traffic Safety, 2012), pp. 1– 8 3 Expert Panel on Driver Fatigue and Sleepiness: ' Drowsy driving and automobile crashes' ( National Center for Sleep Disorders Research/National Highway Traffic Safety Authority, Washington, DC, 1997) 4Cheng B. Zhang W. Lin Y. et al.: 'Driver drowsiness detection based on multisource information', Hum. Factors Ergon. Manuf. Serv. Ind., 2012, 22, (5), pp. 450– 467 5Berka C. Levendowski D. Lumicao M. et al.: 'EEG correlates of task engagement and mental workload in vigilance, learning, and memory tasks', Aviat. Space Environ. Med., 2007, 78, (Suppl. 1), pp. B231– B244 6Tran Y. Wijesuriya N. Tarvainen M. et al.: 'The relationship between spectral changes in heart rate variability and fatigue', J. Psychophysiol., 2009, 23, (3), pp. 143– 151 7Peiris M. Jones R. Davidson P. et al.: 'Frequent lapses of responsiveness during an extended visuomotor tracking task in non-sleep-deprived subjects', J. Sleep Res., 2006, 15, (3), pp. 291– 300 8Johnson R. Popovic D. Olmstead R. et al.: 'Drowsiness/alertness algorithm development and validation using synchronized EEG and cognitive performance to individualize a generalized model', Biol. Psychol., 2011, 87, (2), pp. 241– 250 9Eoh H. Chung M. Kim S.: 'Electroencephalographic study of drowsiness in simulated driving with sleep deprivation', Int. J. Ind. Ergon., 2005, 35, (4), pp. 307– 320 10Eichele T. Debener S. Calhoun V. et al.: 'Prediction of human errors by maladaptive changes in event-related brain networks', Proc. Natl. Acad. Sci., 2008, 105, (16), pp. 6173– 6178 11Jap B. Lal S. Fischer P.: 'Comparing combinations of EEG activity in train drivers during monotonous driving', Expert Syst. Appl., 2011, 38, (1), pp. 996– 1003 12Jap B. Lal S. Fischer P. et al.: 'Using EEG spectral components to assess algorithms for detecting fatigue', Expert Syst. Appl., 2009, 36, (2), pp. 2352– 2359 13Akerstedt T. Gillberg M.: 'Subjective and objective sleepiness in the active individual', Int. J. Neurosci., 1990, 52, (1–2), pp. 29– 37 14Stern M. Jerome E.: ' Atlas of EEG patterns' ( Lippincott Williams & Wilkins, Philadelphia, 2005) 15Torsvall L.: 'Sleepiness on the job: continuously measured EEG changes in train drivers', Electroencephalogr. Clin. Neurophysiol., 1987, 66, (6), pp. 502– 511 16Gillberg M. Kecklund G. Åkerstedt T. et al.: 'Sleepiness and performance of professional drivers in a truck simulator—comparisons between day and night driving', J. Sleep Res., 1996, 5, (1), pp. 12– 15 17Lin T. Ko W. Chung F. et al.: 'Adaptive EEG-based alertness estimation system by using ICA-based fuzzy neural networks', IEEE Trans. Circuits Syst. I, Regul. Pap., 2006, 53, (11), pp. 2469– 2476 18Lin T. Wu C. Jung P. et al.: 'Estimating driving performance based on EEG spectrum analysis', EURASIP J. Appl. Signal Process., 2005, 2005, pp. 3165– 3174 19Stewart D.: 'A platform with six degrees of freedom', Proc. Inst. Mech. Eng., 1965, 180, (1), pp. 371– 386 20Lin C. Ko W. Chen A. et al.: 'EEG-based cognitive state monitoring and prediction by using the self-constructing neural fuzzy system'. Proc. 2010 IEEE Int. Symp. IEEE Circuits and Systems (ISCAS), 2010, pp. 2287– 2290 21Li G. Wang W. Li E. et al.: 'Effectiveness of flashing brake and hazard systems in avoiding rear-end crashes', Adv. Mech. Eng., 2014, 6, pp. 642– 647 22Liao Y. Li S. Wang W. et al.: 'Detection of driver cognitive distraction: a comparison study of stop-controlled intersection and speed-limited highway', IEEE Trans. Intell. Transp. Syst., 2016, 17, (6), pp. 1628– 1637 23Duan J. Li R. Hou L. et al.: 'Driver braking behavior analysis to improve autonomous emergency braking systems in typical Chinese vehicle-bicycle conflicts', Accident Anal. Prev., 2017, 108, pp. 74– 82 24Moller J. Kayumov L. Bulmash L. et al.: 'Simulator performance, microsleep episodes, and subjective sleepiness: normative data using convergent methodologies to assess driver drowsiness', J. Psychosom. Res., 2006, 61, (3), pp. 335– 342 25Gupta R. Mehra R.: 'Power spectrum estimation using Welch method for various window techniques', Int. J. Sci. Res. Eng. Technol., 2013, 2, (6), pp. 389– 392 26Zhao C. Zhao M. Liu J. et al.: 'Electroencephalogram and electrocardiograph assessment of mental fatigue in a driving simulator', Accident Anal. Prev., 2012, 45, pp. 83– 90 27Wierwille W. Ellsworth L.: 'Evaluation of driver drowsiness by trained raters', Accident Anal. Prev., 1994, 26, (5), pp. 571– 581 28Akerstedt T. Peters B. Anund A. et al.: 'Impaired alertness and performance driving home from the night shift: a driving simulator study', J. Sleep Res., 2005, 14, (1), pp. 17– 20 29George F.: 'Vigilance impairment: assessment by driving simulators', Sleep, 2000, 23, (Suppl. 4), pp. S115– S118 30Abtahi F. Anund A. Fors C. et al.: 'Association of drivers' sleepiness with heart rate variability: a pilot study with drivers on real roads'. EMBEC & NBC 2017, 2017, pp. 149– 152 31Belyavin A. Wright A.: 'Changes in electrical activity of the brain with vigilance', Electroencephalogr. Clin. Neurophysiol., 1987, 66, (2), pp. 137– 144 32Lal K. Craig A.: 'Driver fatigue: electroencephalography and psychological assessment', Psychophysiology, 2002, 39, (3), pp. 313– 321 33Melia U. Guaita M. Vallverdú M. et al.: 'Correntropy measures to detect daytime sleepiness from EEG signals', Physiol. Meas., 2014, 35, (10), pp. 2067– 2083 34Miyashita T. Ogawa K. Itoh H. et al.: 'Spectral analyses of electroencephalography and heart rate variability during sleep in normal subjects', Auton. Neurosci., 2003, 103, (1–2), pp. 114– 120 35Shen Q. Ong J. Li P. et al.: 'A feature selection method for multilevel mental fatigue EEG classification', IEEE Trans. Biomed. Eng., 2007, 54, (7), pp. 1231– 1237 36Filtness J. Reyner L. Horne J.: 'Driver sleepiness-comparisons between young and older men during a monotonous afternoon simulated drive', Biol. Psychol., 2012, 89, (3), pp. 580– 583 Citing Literature Volume12, Issue5June 2018Pages 345-350 FiguresReferencesRelatedInformation