SSRN Id4570901
SSRN Id4570901
SSRN Id4570901
we
Driver Drowsiness Detection Using Ocular Artefacts
and Alpha Band Power Variations in EEG Signal
vie
Department of Electrical and Electronic Engineering
Chittagong University of Engineering and Technology
Chattogram-4349, Bangladesh
Contact Number: +880-1713109985
Email: mrthossain@cuet.ac.bd
*Corresponding author
re
Tanzila Islam
Department of Electrical and Electronic Engineering
Chittagong University of Engineering and Technology
er
Chattogram-4349, Bangladesh
Email: u1702019@student.cuet.ac.bd
Arpita Mazumder
pe
Department of Electrical and Electronic Engineering
Chittagong University of Engineering and Technology
Chattogram-4349, Bangladesh
Email: u1702094@student.cuet.ac.bd
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
Abstract: Drowsy driving is one of the leading causes of increasing number of
traffic accidents with numerous deaths, injuries, and monetary losses every
year. This can be significantly reduced if a system is developed to detect driver
drowsiness at an early stage. This paper presents an approach utilizing the eye
blinking information extracted from the ocular artefact and the alpha band
power variations from the same Electroencephalogram (EEG) signal to detect
driver drowsiness. The band-wise sorting of EEG data exhibits alpha wave
vie
(frequency range: 8–13 Hz) band power rise when a driver feels drowsy. The
ocular artefact, which is otherwise considered as noise in EEG signal, is
extracted to provide eye blinking information to identify fatigue. The research
also associates the prefrontal brain region as better responsive for EEG signal
acquisition both for observing the alpha band power shift and collecting the eye
blinking data easily with only two electrodes. Using three selected features with
simple algorithm drowsiness detection is accomplished with greater success as
both methods are utilized complementary with each of its distinguishing
re
characteristics.
methods fall into three categories: those based on behavioural, vehicular, and
physiological parameters (Zheng et al., 2022). The most common behavioural parameters
used that evaluate the rate of eye or mouth closure are PERCLOS (percentage of eye
closure) and Mouth Aspect Ratio. When drowsiness sets in, these rates tend to exceed the
threshold value. Though these systems are non-invasive and rely on recorded video to
rin
locate facial landmarks, there are significant risks of failure for poor illumination and
vehicular movement on uneven roads, etc. (Nguyen, Chew and Demidenko, 2015; Mittal
et al., 2016; Omidyeganeh et al., 2016; Reddy et al., 2017; Jabbar et al., 2018). The
vehicular parameters based on the signal structure provided by the drivers, such as the
vehicle's wheel angular velocity and steering angles are used to identify notable patterns
ep
(Lawoyin et al., 2015; Arefnezhad et al., 2020). The layout of the route, weather, or
changes to grip pressure and speed can harm pattern changes that indicate drowsiness. On
the contrary, physiological indicators based on Electroencephalogram (EEG),
Electrocardiogram (ECG), and Electrooculogram (EOG) signals can be more useful in
identifying drowsiness as they appear in the body at the very beginning of feeling
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
tiredness (Fujiwara et al., 2019; Majumder et al., 2019; Jia, Xiao and Ji, 2023). EEG
signal reflects the neurobehavioral pattern which is influenced by drowsiness or sleep
(Gurrala, Yarlagadda and Koppireddi, 2021; Min et al., 2021; Shen et al., 2021; Tuncer,
Dogan and Subasi, 2021). Hence, analysing EEG signals for drowsiness detection gained
much attention from researchers. EEG signal is the electric activity of a large number of
neurons in the brain, recorded with electrodes applied to the scalp’s surface. These
vie
signals are generally categorized as delta, theta, alpha, beta and gamma based on signal
frequencies ranging from 0.1 Hz to more than 100 Hz (Reddy et al., 2017; Turkoglu et
al., 2021). Each of these brain waves in the active phase shows certain patterns of activity
related to a person’s mental health and behaviour (Murugappan et al., 2009; Elgandelwar
and Bairagi, 2021). These rhythms’ band wave strengths can change while a person
performs various activities and this fact can be analysed to evaluate fatigue along with the
re
psychological state (Rizon et al., 2020).
EEG signals have very small amplitudes and hence are prone to contamination by
different artefacts such as baseline wandering, power line noise, eye movements,
Electromyogram (EMG) disturbance, and Electrocardiogram (ECG) disturbance
(Sheoran, Kumar and Chawla, 2015). These artefacts are regarded as noise and are
removed during EEG pre-processing. Ocular artefact originates by eye blinking and
er
movement, which spreads throughout the entire scalp. Thus the information regarding
eye blinking and the level of eye closure can be obtained through the ocular artefact
extraction from the EEG signal. If the ocular artefacts from the EEG signal can be used
for drowsiness detection, the acquisition of EOG signals using a distinct setup by
pe
positioning electrodes close to the eyes will be no longer required. This will reduce the
complexity and improve the efficacy of the system.
According to the Sleep Foundation in Washington, DC, USA (Summer, 2022), the
fundamental brain wave pattern that emerges during the transition from wakefulness to
sleep is known as alpha waves. As a person participates in more focused mental activity,
the generation of alpha wave starts to decrease and beta wave predominates. Just before
falling asleep, alpha waves start to appear more frequently before being replaced by theta
ot
waves. The alpha band power in specific brain regions is found to increase when the
eyelids are closed and practically disappear when the eyes are opened (Kim et al., 2012;
Ben Dkhil et al., 2017). C. K. A. Lim et al. proposed a method for identifying fatigue
using alpha band power with a single channel of observation from the prefrontal region
tn
(Lim, Chia and Chin, 2014). In (Gharagozlou et al., 2015), variations in alpha and delta
band power in the parietal area are evaluated and an increase in alpha power during long-
distance driving with a decrease in consciousness is observed.
Recent studies have shown that frontal cortical activity during sleep is characterized
by high voltage and slow brain waves, especially in the prefrontal cortex (Muzur, Pace-
rin
Schott and Hobson, 2002). This cortex is a part of the brain that is easiest to notice
changes in band power transitions. The basic concept behind this approach is that alpha
activity shifts gradually from the frontal lobe to the occipital region during different
phases of sleep (Van Hal et al., 2014; Wang et al., 2021). Therefore, using the frontal
area to identify alpha band power alterations can provide earlier symptoms than other
ep
regions.
Through the utilization of certain characteristics from the ocular extraction, S.
Tarafder et al. revealed great potential for distinguishing between alert and drowsy states
(Tarafder et al., 2022). The BLINKER algorithm was used to extract 25 features from the
ocular artefact of the EEG data from all regions of the brain. The high number of features
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
increased the computational complexity of their work. Moreover, information is absent
regarding how the features change when drowsiness sets in.
Based on this concept, M. Shahbakhti et al. evaluated the ocular artefact’s function as
noise and eye information (Shahbakhti et al., 2022). Prefrontal cortex EEG recordings
were used to assess the changes in filtered and raw signals as a person becomes sleepy.
During noise removal from EEG, they removed only the EOG signal. But they did not
vie
mention the ECG, EMG, and other artefacts.
Another study by Shahbakhti et al. proposed a fusion approach that used all brain
waves and gathered 8 features (power ratio and entropy related) from each of the five
bands. Hence, a total of 40 features from EEG were collected. The blinking signals were
also utilized to measure the blink rate, average blink amplitude, and blink distance
(Shahbakhti et al., 2023). Therefore, in total 43 features were used here. Using
re
neighbourhood component analysis (NCA) the number of features was reduced to 18.
They used entropy and fractal dimension-based features which are computationally
expensive. Moreover, the number of features is more which made the model complex.
Although the prefrontal region was studied, no specific reasoning compared to other
brain regions was presented. This study did not include the removal of any other bodily
artefacts such as ECG, EMG, etc. from the raw EEG signal.
er
No research was conducted to observe the simultaneous change in ocular signals and
single band power such as the alpha band to detect drowsiness. Merging ocular signal
features and alpha band power to form reduced numbers of biomarkers can lead to a
simpler detection scheme.
pe
The scalp area with the highest level of brain activity and where it seems more
feasible than other regions must be found to do the band power spectrum analysis.
Different bands and brain regions respond to drowsiness in various manners. It is possible
to detect functional changes more quickly and with less processing complexity through
fewer placements of electrodes and getting a better band response in specific brain
regions. The alpha wave can be used as an indicator to determine when drowsiness
begins because it develops between the awake and sleepy states. To increase the level of
ot
establish a person's status as he switches from awake to drowsy state. It will be a more
precise early-stage indicator than detection from the video recording (Nguyen, Chew and
Demidenko, 2015; Mittal et al., 2016; Omidyeganeh et al., 2016; Reddy et al., 2017;
Jabbar et al., 2018) if physiological parameters can be employed to collect data on
blinking. Ocular artefact, also known as EEG signal contamination, raises the chance that
rin
EEG signals may carry information about our eye dynamics. Using the BLINKER
Algorithm (Agarwal and Sivakumar, 2019), it is possible to extract the EOG that is
typically regarded as EEG contamination. Extracting EOG from the same EEG signal, the
possibility of detecting drowsiness can be increased by obtaining the blinking features
and how they fluctuate as a person starts dozing off.
ep
This paper introduces an approach that involves assessing alpha band power variation
and extracting features of ocular artefacts both from a single EEG channel. Firstly, a
particular region of the scalp has been selected where the alpha band power appeared
more responsive than other regions which lead to a lower number of electrodes used for
detection. The power spectrum analysis is performed for each participant to observe the
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
power variation between the alert and drowsy states. Then, some relevant features of
ocular artefacts have been extracted for the prediction of a higher blinking rate in drowsy
conditions. Eventually, utilizing two distinct methods from the same participants, we
investigated how band power variation and ocular artefact features behave in drowsy
individuals before proposing an approach based on our findings. Finally, the findings
from both methods are combined to make a simpler algorithm which is described in the
vie
next section.
The major contributions of this study are as follows:
re
Unlike other studies, in our study, ECG, EMG, and other artefacts are
removed before analysing the EEG.
Instead of using a machine learning model, a simplified explainable
algorithm is used.
To the best of our knowledge, there is no approach being proposed that
er
combined alpha band power (instead of using all bands) and ocular artefact
feature analysis simultaneously to detect drowsiness. We apply this approach
in our study.
pe
2 Materials and Methods
This section is divided into four parts. The dataset and the study's algorithms are
discussed in the first two parts. The remaining parts describe the methodologies which
focused on the alpha band's power spectral density (PSD) analysis and ocular artefact
analysis for driver drowsiness detection. The entire procedure as illustrated by the flow
ot
2.1 Dataset
tn
We used the open-access dataset available from (Min, Wang and Hu, 2017) for our
research. From the dataset, we used the EEG data of twelve healthy young individuals
from Jiangxi University of Technology, China, who participated in a highway driving
simulation experiment for driver fatigue detection. The EEG data of the participants were
collected using a 40-channel Neuroscan amplifier in two stages. The first stage involved
driving for a short period (20 min). The final five minutes of EEG signals were collected
rin
and considered as the normal state. The second stage involved continuous driving for a
long duration (40 - 100 min) until the individuals nodded off. The last five minutes of
EEG signals were recorded and labelled as the drowsy state.
ep
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
vie
re
er
pe
ot
2.2 Algorithms
2.2.1 Independent component analysis (ICA)
EEG data is a mixture of source and artefact signals. Therefore, the quality of the EEG
rin
data can be raised by eliminating the artefact signals. The artefact signals from the EEG
data are eliminated in EEGLab by using the ICA algorithm. This method first takes a
linear combination of the signals coming from several electrodes or sensors. The
combined signals are then broken down into numerous separate brain and non-brain
components. This allows the elimination of the non-brain components as necessary.
ep
2.2.2 BLINKER
The ocular artefact in EEG data is normally considered as noise, but it can also reveal
useful information. The BLINKER algorithm of EEGLab, which works on the MATLAB
platform, is used to extract eye blink artefacts and their features like blink rate, blink
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
duration, blink amplitude-velocity ratios, etc. from EEG signal (Kleifges et al., 2017).
The technique involves first extracting ocular artefacts from the raw EEG data and then
putting them through a 1-20 Hz Finite Impulse Response (FIR) filter. After that, to
exclude rapid eye movements, this method only takes into account eye blinks that have a
high amplitude, a maximum duration of 50 milliseconds, and are spaced by at least 50
milliseconds. Finally, it provides 25 features for the selected blinks.
vie
2.3 Alpha Band Power Spectrum Analysis
2.3.1 EEG pre-processing
The pre-processing of raw EEG data is done by filtering the signal with a 0.5-40 Hz FIR
filter to remove line noise and dc offset. However, the data also contains some artefacts
re
which contaminate the EEG signal and become a hinder to obtaining proper brain signal.
The removal of artefacts from an EEG signal can be accomplished in several ways.
However, unlike other techniques that can only deal with certain kinds of artefacts, ICA
algorithms can manage all sorts of EEG artefacts (Jiang, Bian and Tian, 2019).
The filtered EEG data is shown in Figure 2 before artefact removal. To separate the
er
filtered data into the multiple ICA components as depicted in Figure 3, the Extended ICA
approach from the EEGLab toolbox is applied. This algorithm can properly identify the
necessary components and estimate them all at once (Matsuda and Yamaguchi, 2022).
Besides, it performs ICA decomposition on psychophysiological data and can also detect
considerable line noise. EEGLab removes artefact components by deciding on a rejection
pe
range that conveys us the flexibility to adjust the range and enables us to recognize a
component as an artefact if it falls within the given range. Applying this function, except
for the eye component, the other artefact components (muscle, heart, channel noise, etc.)
are eliminated by setting a rejection range of 0.9 to 1. Due to acute data contamination of
the eye component, it is eliminated by setting the rejection range from 0.5 to 1. Figure 4
displays the EEG data after artefact removal.
ot
tn
rin
ep
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
vie
re
er
pe
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
2.3.2 Power spectrum analysis
Following the removal of artefacts, the alpha band power is observed in EEGLab using a
few selected channels from different scalp regions, including Fp1 from the prefrontal
cortex, F4, and F7 from the frontal region, P3 from the parietal region, O1, and O2 from
the occipital region. EEGLab uses Welch’s method (Li et al., 2010; Saidatul et al., 2011)
vie
and the Fast Fourier Transform (FFT) method to display the power spectrum of EEG data
from these scalp-mounted electrodes. An example of power spectrum analysis of EEG
data during the drowsy state at the selected scalp regions is displayed in Figure 5. The
re
er
(a) Fp1 channel (b) F4 channel
pe
ot
Figure 5 An illustration of scalp region selection using different regions of a participant. The
analysis of the entire signal activity over 200 segments (duration of each segment is 1.5 seconds) is
shown in the 'Continuous data' portion in the top right corner of each figure
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
prefrontal region has high alpha band power for a majority of the participants when they
started to nod off. Consequently, the prefrontal scalp region is chosen here for alpha band
power analysis.
Finally, for each participant, how the alpha band power varied from a wakeful to a
sleepy state is analysed via both Fp1 and Fp2 electrodes in the prefrontal region. A
change in alpha band power in the prefrontal area is illustrated in Figure 6.
vie
re
er
(a) Normal state via Fp1 (b) Drowsy state via Fp1
pe
ot
tn
(c) Normal state via Fp2 (d) Drowsy state via Fp2
Figure 6 Alpha band power spectrum of a participant in the normal and drowsy state in the
rin
prefrontal region (a) and (b) demonstrate Fp1 electrode (c) and (d) demonstrate Fp2 electrode
or not the driver is drowsy. In ocular artefact analysis, at first, a set of potential eye blink
components with 25 blink-related features are extracted from EEG signals of both normal
and drowsy individuals using the BLINKER algorithm for further investigation. Certain
ocular features can be chosen to simplify the method to identify participant fatigue.
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
2.4.1 Selecting eye-blink features
Among 25 features provided by the BLINKER algorithm, we have chosen 3 features for
our study which are correlated to EEG band power and can act as good indicators of
drowsiness (Ftouni et al., 2013). These features are Blink rate, Positive Amplitude
Velocity Ratio (pAVR) and Negative Amplitude Velocity Ratio (nAVR). These are
vie
explained below:
Blink rate: This feature denotes the number of blinks a person has in a minute. A
drowsy person seems to have a greater number of long-duration eye blinks which makes
the blinks more noticeable in the extracted signal. The situation is quite different for
people who are awake.
pAVR: pAVR is the ratio of the signal’s maximum amplitude to eye closure velocity
re
ratio. The rate of ocular closure is higher in alert people due to their quick blinking,
whereas it is lower in drowsy people. It makes tired people have higher pAVR than alert
people.
nAVR: nAVR is the ratio of the signal’s maximum amplitude to the eye-open
velocity ratio. Similar to pAVR, it rises in drowsy individuals and falls in normal ones.
er
2.4.2 Observation of normal and drowsy state of participants
We studied how these three eye blink characteristics changed from a wakeful to a drowsy
condition in the prefrontal region for each of our subjects as shown in Figure 7. Finally,
the features that most closely matched our assumption in the case of maximum
pe
participants are used in our drowsiness detection study.
ot
tn
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
3 Results and Discussion
Every area of the brain that is currently utilized for tracking changes in band powers has
been explored, such as the prefrontal, frontal, parietal, and occipital regions. We have
chosen fewer electrodes than usual because the band power transitions generated by
channels from similar locations are quite alike. For each participant, the alpha responses
vie
from the chosen channels are stated in Table 1.
Electrode
Fp1 F4 F7 P3 O1 O2
/Participant
Person 1 3.213 2.733 2.584 0.7132 3.473 -2.639
re
Person 2 6.98 5.459 5.97 7.597 13.87 15.36
For the majority of individuals, the alpha band power has a higher value in the
tn
prefrontal (Fp1) region than in other regions. Figure 8 shows the percentage of the alpha
response that increased in the twelve participants’ different brain regions.
Although the prefrontal region doesn't always display the highest value for each
situation, there is never a case when the value is negative or insufficient to be taken into
account. Above all, this is the only hairless area of the human head from where the signal
rin
can be extracted easily and comfortably with less probability of signal distortion. Thus,
due to responsiveness and the good readability of signals, two (Fp1, Fp2) prefrontal
electrodes are chosen for observing drowsiness to ensure that, in case one electrode is
unable to read the signal, the other can do so. This increases the reliability of the system.
Following the selection of the prefrontal region as suitable for drowsiness detection,
ep
the variations in alpha band power in Fp1 and Fp2 channels are observed for all
participants when they shift from normal to drowsy state. For each individual, the alpha
band power values in both awake and sleepy states are plotted in Figure 9.
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
vie
re
Figure 8 The percentages of increased alpha wave responses in different brain regions during the
transition from normal to a drowsy state
The alpha band power during the transition from normal to drowsy condition is
expected to be raised due to the dominance of alpha waves in the prefrontal area. Our
er
study likewise showed that for the majority of the individuals, in FP1 and FP2 prefrontal
regions, the alpha band power is higher during the sleepy stage than that during the
normal stage. The percentages of the results are plotted in Figure 10. The pie chart
reveals that for 75% of the participants, both channels in the prefrontal region, Fp1 and
Fp2, could successfully detect drowsiness by showing increased alpha band power in the
pe
drowsy state. For 8% of cases, either Fp1 or Fp2 could detect drowsiness accurately. The
unreliable 17% of cases may still be resolved to identify drowsiness if the alternate
approach via the ocular artefact analysis can be employed to find characteristics from the
same EEG signal.
ot
tn
rin
ep
Figure 9 Alpha band power in the normal and drowsy state for each participant in the prefrontal
region (Fp1 and Fp2 channels)
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
vie
re
Figure 10 Summary of alpha band power responses in the prefrontal region (Fp1 and Fp2
channels)
For ocular artefact analysis, we have chosen three specific features, including blink
rate, pAVR, and nAVR, along with analysing their variability to employ them as a way to
er
detect drowsiness. Each participant is expected to have greater numbers of long-duration
eye blinks per minute with higher pAVR and nAVR values in the drowsy condition. For
all participants, the percentage variation of each selected feature during the transition
from normal to drowsy state is shown in Table 2.
pe
Table 2 The change in outcomes to identify three ocular features’ responses
Person 2
Person 7 -30% 2% 5%
Person 10 47% 2% 1%
Person 12
The Table shows that for seven out of twelve participants, both the blink rate and the
nAVR features indicate an increase in drowsiness in their cases. As they both have a
chance of being inaccurate due to small errors, these features can be used as
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
complementary drowsiness indicators. For the pAVR feature, it is unclear whether it has
an increasing or decreasing tendency. Hence for further analysis, the pAVR feature is not
considered. For 3 subjects the BLINKER algorithm failed to extract Blink rate, nAVR,
and pAVR features. This indicates the deficiency of the BLINKER algorithm. Despite
this deficiency, our method still works because it also tests alpha band power as an
alternative.
vie
A summary of the results of power spectrum analysis and ocular artefact analysis for
drowsiness detection of twelve individual participants is listed in Table 3. Analysis of the
results of alpha band power spectrum analysis shows that for ten out of twelve subjects,
the drowsiness is detected properly therefore the alpha band power increased in the
drowsy state. Analysis of the results of ocular artefact feature analysis showed that for
seven out of twelve subjects, the drowsiness is detected properly. When the change in
re
alpha band powers cannot be detected by spectrum analysis, the ocular artefact analysis
may help in the identification of the state through various indicators, such as blinks/min,
and nAVR, as we have seen in our participants.
Table 3 Different results of the two methods showing significant alterations in the drowsy state.
Participants
er Power spectrum
analysis
A rise in alpha band
Ocular artefact analysis
Person 3
Person 4
Person 5
ot
Person 6
Person 7
tn
Person 8
Person 9
Person 10
rin
Person 11
Person 12
Combining the two methods, the potential for identifying a drowsy condition can be
ep
significantly enhanced than the individual methods. The same EEG signal is employed by
both methods, and their respective inputs are denoised EEG and extracted EOG from
EEG. We have focussed on overcoming the research gaps in earlier works on drowsiness
detection which are summarized in Table 4.
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
Table 4 An overview of previous works and our suggested approach to detect drowsiness
vie
(KumarAhirwal and D londhe, 2012), (Kim et Low-frequency Always approach was
al., 2012), (Lim, Chia and Chin, 2014), (Van waves, filtered out developed
Hal et al., 2014), (Gharagozlou et al., 2015), especially alpha and treated
(Ben Dkhil et al., 2017), (Majumder et al., waves as noise
2019), (Wang et al., 2021)
(Tarafder et al., 2022), (Shahbakhti et al., Not used Used. No fusion
re
2022) 25 features approach was
were developed
extracted
(Shahbakhti et al., 2023) Used. Used. Used
16 features from 2 features
5 bands were were used
used
er
Proposed Research Used.
1 feature is
Used.
2 features are
Used.
The same
used. used EEG signal is
(A specific used in both
brain region is methods.
pe
utilized to
observe only
alpha band
power)
Our proposed work solved the shortcomings of the previous methods. Despite taking
all five brain waves, we have only considered the alpha wave for band power analysis. In
ot
addition, two features from ocular artefact are used. In total, we have used only 3 features
which resulted in a less complex system. The same EEG signal is used in both methods.
The channel selection (Fp1 and/or Fp2) is done considering both the driver's comfort and
quality signal extraction. ECG, EMG, and other artefacts are removed before analysing
tn
the EEG. Instead of using a machine learning model, a simplified explainable algorithm
is used.
Although the proposed method works well, it has the following limitations:
As the data of this study were gathered in a simulated driving environment,
participants could not experience driving in practical situations. To study the feasibility
of the proposed scheme, it should be tested in real-life scenarios.
ep
In this study, the BLINKER algorithm was unable to extract ocular indices from the
prefrontal region in the case of 3 participants.
The ICA method could not effectively remove all the artefact components from the
EEG data.
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
5 Conclusion
This research is conducted to develop an EEG signal-based simple strategy for detecting
driver drowsiness using only three features with a simplified algorithm. The scheme
comprises the combination of two methods - drowsiness detection by investigating the
alpha wave band power variations in the filtered EEG signal and also by observing the
changes in the 2 selected eye blink features extracted from the ocular artefact that is
vie
mixed with the same EEG signal as noise. The selection of channel at the prefrontal
region conforms to both good signal acquisition as well as driver's ease. Drowsiness is
identified for every participant when the results are combined from each method. Taking
into account the findings and comparison with earlier studies, it can be concluded that the
suggested approach achieves the intended goals and can identify drowsiness. The scheme
should be tested in practical driving situations for further technical evaluation.
re
References
Agarwal, M. and Sivakumar, R. (2019) ‘Blink: A Fully Automated Unsupervised Algorithm for
er
Eye-Blink Detection in EEG Signals’, in 2019 57th Annual Allerton Conference on
Communication, Control, and Computing, Allerton 2019. IEEE, pp. 1113–1121. Available at:
https://doi.org/10.1109/ALLERTON.2019.8919795.
Arefnezhad, S. et al. (2020) ‘Applying deep neural networks for multi-level classification of driver
drowsiness using Vehicle-based measures’, Expert Systems with Applications, 162, p. 113778.
pe
Available at: https://doi.org/10.1016/j.eswa.2020.113778.
CDC (2023) Global Road Safety | Transportation Safety | Injury Center | CDC, Road Traffic
Injuries and Deaths—A Global Problem. Available at: https://www.cdc.gov/injury/features/global-
road-safety/index.html (Accessed: 18 June 2023).
Ben Dkhil, M. et al. (2017) ‘Towards an automatic drowsiness detection system by evaluating the
alpha band of EEG signals’, in SAMI 2017 - IEEE 15th International Symposium on Applied
ot
Machine Intelligence and Informatics, Proceedings. IEEE, pp. 371–376. Available at:
https://doi.org/10.1109/SAMI.2017.7880336.
Elgandelwar, S.M. and Bairagi, V.K. (2021) ‘Power analysis of EEG bands for diagnosis of
Alzheimer disease’, International Journal of Medical Engineering and Informatics, 13(5), pp. 376–
385. Available at: https://doi.org/10.1504/IJMEI.2021.117728.
tn
Ftouni, S. et al. (2013) ‘Temporal dynamics of ocular indicators of sleepiness across sleep
restriction’, Journal of Biological Rhythms, 28(6), pp. 412–424. Available at:
https://doi.org/10.1177/0748730413512257.
Fujiwara, K. et al. (2019) ‘Heart Rate Variability-Based Driver Drowsiness Detection and Its
Validation With EEG’, IEEE Transactions on Biomedical Engineering, 66(6), pp. 1769–1778.
rin
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
Jabbar, R. et al. (2018) ‘Real-time Driver Drowsiness Detection for Android Application Using
Deep Neural Networks Techniques’, Procedia Computer Science, 130, pp. 400–407. Available at:
https://doi.org/10.1016/j.procs.2018.04.060.
Jia, H., Xiao, Z. and Ji, P. (2023) ‘End-to-end fatigue driving EEG signal detection model based on
improved temporal-graph convolution network’, Computers in Biology and Medicine, 152, p.
106431. Available at: https://doi.org/10.1016/j.compbiomed.2022.106431.
vie
Jiang, X., Bian, G. Bin and Tian, Z. (2019) ‘Removal of artifacts from EEG signals: A review’,
Sensors (Switzerland), 19(5), p. 987. Available at: https://doi.org/10.3390/s19050987.
Kim, D. et al. (2012) ‘Detection of drowsiness with eyes open using EEG-based power spectrum
analysis’, in Proceedings - 2012 7th International Forum on Strategic Technology, IFOST 2012.
IEEE, pp. 1–4. Available at: https://doi.org/10.1109/IFOST.2012.6357815.
Kleifges, K. et al. (2017) ‘BLINKER: Automated extraction of ocular indices from EEG enabling
re
large-scale analysis’, Frontiers in Neuroscience, 11(FEB). Available at:
https://doi.org/10.3389/fnins.2017.00012.
KumarAhirwal, M. and D londhe, N. (2012) ‘Power Spectrum Analysis of EEG Signals for
Estimating Visual Attention’, International Journal of Computer Applications, 42(15), pp. 34–40.
Available at: https://doi.org/10.5120/5769-7993.
Lawoyin, S. et al. (2015) ‘Evaluating the efficacy of an accelerometer-based method for drowsy
er
driving detection’, International Journal of Vehicle Safety, 8(2), p. 165. Available at:
https://doi.org/10.1504/IJVS.2015.068691.
Leng, L.B., Giin, L.B. and Chung, W.Y. (2015) ‘Wearable driver drowsiness detection system
based on biomedical and motion sensors’, in 2015 IEEE SENSORS - Proceedings. IEEE, pp. 1–4.
pe
Available at: https://doi.org/10.1109/ICSENS.2015.7370355.
Li, N. et al. (2010) ‘Analysis of EEG signals during acupuncture using spectral analysis
techniques’, 2010 IEEE International Conference on Automation and Logistics, ICAL 2010, pp.
391–395. Available at: https://doi.org/10.1109/ICAL.2010.5585313.
Lim, C.K.A., Chia, W.C. and Chin, S.W. (2014) ‘A mobile driver safety system: Analysis of
single-channel EEG on drowsiness detection’, in 2014 International Conference on Computational
Science and Technology, ICCST 2014. IEEE, pp. 1–5. Available at:
ot
https://doi.org/10.1109/ICCST.2014.7045175.
Luo, H. et al. (2019) ‘Research on fatigue driving detection using forehead EEG based on adaptive
multi-scale entropy’, Biomedical Signal Processing and Control, 51, pp. 50–58. Available at:
https://doi.org/10.1016/j.bspc.2019.02.005.
tn
Majumder, S. et al. (2019) ‘On-board drowsiness detection using EEG: Current status and future
prospects’, in IEEE International Conference on Electro Information Technology. IEEE, pp. 483–
490. Available at: https://doi.org/10.1109/EIT.2019.8833866.
Matsuda, Y. and Yamaguchi, K. (2022) ‘Unique estimation in EEG analysis by the ordering ICA’,
PLoS ONE, 17(10 October), p. e0276680. Available at:
rin
https://doi.org/10.1371/journal.pone.0276680.
Min, J. et al. (2021) ‘Driver fatigue detection based on prefrontal EEG using multi-entropy
measures and hybrid model’, Biomedical Signal Processing and Control, 69, p. 102857. Available
at: https://doi.org/10.1016/j.bspc.2021.102857.
Min, J., Wang, P. and Hu, J. (2017) ‘Driver fatigue detection through multiple entropy fusion
analysis in an EEG-based system’, PLoS ONE, 12(12), p. e0188756. Available at:
ep
https://doi.org/10.1371/journal.pone.0188756.
Mittal, A. et al. (2016) ‘Head movement-based driver drowsiness detection: A review of state-of-
art techniques’, in Proceedings of 2nd IEEE International Conference on Engineering and
Technology, ICETECH 2016. IEEE, pp. 903–908. Available at:
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
https://doi.org/10.1109/ICETECH.2016.7569378.
Murugappan, M. et al. (2009) ‘An investigation on visual and audiovisual stimulus based emotion
recognition using EEG’, International Journal of Medical Engineering and Informatics, 1(3), pp.
342–356. Available at: https://doi.org/10.1504/IJMEI.2009.022645.
Muzur, A., Pace-Schott, E.F. and Hobson, J.A. (2002) ‘The prefrontal cortex in sleep’, Trends in
Cognitive Sciences, 6(11), pp. 475–481. Available at: https://doi.org/10.1016/S1364-
vie
6613(02)01992-7.
Nguyen, T.P., Chew, M.T. and Demidenko, S. (2015) ‘Eye tracking system to detect driver
drowsiness’, in ICARA 2015 - Proceedings of the 2015 6th International Conference on
Automation, Robotics and Applications. IEEE, pp. 472–477. Available at:
https://doi.org/10.1109/ICARA.2015.7081194.
Omidyeganeh, M. et al. (2016) ‘Yawning Detection Using Embedded Smart Cameras’, IEEE
re
Transactions on Instrumentation and Measurement, 65(3), pp. 570–582. Available at:
https://doi.org/10.1109/TIM.2015.2507378.
Reddy, B. et al. (2017) ‘Real-Time Driver Drowsiness Detection for Embedded System Using
Model Compression of Deep Neural Networks’, in IEEE Computer Society Conference on
Computer Vision and Pattern Recognition Workshops. IEEE, pp. 438–445. Available at:
https://doi.org/10.1109/CVPRW.2017.59.
er
Rizon, M. et al. (2020) ‘Eeg based drowsiness detection using relative band power and short time
fourier transform’, Proceedings of International Conference on Artificial Life and Robotics, 2020,
pp. 323–327. Available at: https://doi.org/10.5954/ICAROB.2020.OS10-2.
Saidatul, A. et al. (2011) ‘Analysis of EEG signals during relaxation and mental stress condition
pe
using AR modeling techniques’, in Proceedings - 2011 IEEE International Conference on Control
System, Computing and Engineering, ICCSCE 2011. IEEE, pp. 477–481. Available at:
https://doi.org/10.1109/ICCSCE.2011.6190573.
Saleem, S. (2022) ‘Risk assessment of road traffic accidents related to sleepiness during driving: a
systematic review’, Eastern Mediterranean Health Journal, pp. 695–700. Available at:
https://doi.org/10.26719/emhj.22.055.
Shahbakhti, M. et al. (2022) ‘Simultaneous Eye Blink Characterization and Elimination from Low-
ot
Channel Prefrontal EEG Signals Enhances Driver Drowsiness Detection’, IEEE Journal of
Biomedical and Health Informatics, 26(3), pp. 1001–1012. Available at:
https://doi.org/10.1109/JBHI.2021.3096984.
Shahbakhti, M. et al. (2023) ‘Fusion of EEG and Eye Blink Analysis for Detection of Driver
tn
Fatigue’, IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31, pp. 2037–
2046. Available at: https://doi.org/10.1109/TNSRE.2023.3267114.
Shen, M. et al. (2021) ‘Multi-source signal alignment and efficient multi-dimensional feature
classification in the application of EEG-based subject-independent drowsiness detection’,
Biomedical Signal Processing and Control, 70, p. 103023. Available at:
https://doi.org/10.1016/j.bspc.2021.103023.
rin
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901
d
we
Tarafder, S. et al. (2022) ‘Drowsiness Detection Using Ocular Indices from EEG Signal’, Sensors,
22(13), p. 4764. Available at: https://doi.org/10.3390/s22134764.
Tuncer, T., Dogan, S. and Subasi, A. (2021) ‘EEG-based driving fatigue detection using multilevel
feature extraction and iterative hybrid feature selection’, Biomedical Signal Processing and
Control, 68, p. 102591. Available at: https://doi.org/10.1016/j.bspc.2021.102591.
Turkoglu, M. et al. (2021) ‘Deep rhythm and long short term memory-based drowsiness detection’,
vie
Biomedical Signal Processing and Control, 65, p. 102364. Available at:
https://doi.org/10.1016/j.bspc.2020.102364.
Wang, H. et al. (2021) ‘Driving Fatigue Recognition with Functional Connectivity Based on Phase
Synchronization’, IEEE Transactions on Cognitive and Developmental Systems, 13(3), pp. 668–
678. Available at: https://doi.org/10.1109/TCDS.2020.2985539.
Zheng, Y. et al. (2022) ‘A new feature selection approach for driving fatigue EEG detection with a
re
modified machine learning algorithm’, Computers in Biology and Medicine, 147, p. 105718.
Available at: https://doi.org/10.1016/j.compbiomed.2022.105718.
er
pe
ot
tn
rin
ep
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4570901