Received October 4, 2019, accepted October 21, 2019, date of publication November 5, 2019, date of current version December

2, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2951028

Drowsiness, Fatigue and Poor Sleep’s Causes and

Detection: A Comprehensive Study
M. AHMAD KAMRAN , MALIK M. NAEEM MANNAN , AND MYUNG YUNG JEONG
Cogno-Mechatronics Engineering, Pusan National University, Busan 46241, South Korea
Corresponding author: Myung Yung Jeong ([email protected])
This work was supported by the National Research Foundation of Korea (NRF) Grant funded by the Korean Government (MSIP)
(No. 2015R1A5A1037668).

ABSTRACT Drowsiness/sleepiness is a serious issue that needs to be addressed for improvement in

the safety of road driving. Past statistical data on road accidents has shown enormous increases in car
crashes due to drowsy/sleepy feelings. This study comprehensively summarizes all aspects of the drowsy
state and its effects during car driving: its symptoms, causes, preventive actions, car accident statistics,
sleep stages, and the behavioral, physiological and neural activation changes occurring during wakefulness
and in the drowsy state. It considers drivers’ behavioral data and corresponding methodologies for its
analysis, the biomedical signals of the human body (including neuronal signals in the forms of electrical
and hemodynamic responses), and their use for drowsiness detection. All of the existing methodologies,
their uses and pros and cons, are comprehensively summarized. A detailed survey of the data published by
neuro-imaging methodology-, physiological signal- and behavioral methodology-based studies in addition
to studies using electro-mechanical installed sensors are statistically and theoretically summarized. Addi-
tionally, the neuronal activity occurring during the drowsy and awake states are analyzed, and the important
contributions of fNIRS, fMRI and EEG in this context are discussed in detail. Differing existing drowsiness-
detection systems installed in popular car brands also are reviewed. Finally, the remaining challenges and
future suggestions for drowsiness-detection systems are summarized as well.

INDEX TERMS Drowsiness, fatigue, functional near-infrared spectroscopy, functional magnetic resonance
imaging, electroencephalography.

I. INTRODUCTION of the human brain’s processing efficiency [8]–[10]. Thus,

Car driving in a drowsy state or with attention distraction during the drowsy stage, there is a load on the human brain
is a serious problem that results in large numbers of cars that renders difficulty in the performance of tasks that would
crashes each year [1], [2]. Such crashes lead to great loss be easily done in the alert stage. The gradual loss in efficiency
of life as well as countless injuries and disabilities [3]. results from malfunction of the cortical network for decision-
The statistical data on car crashes available at the national making [8], [11]. Such situation leads in cutting off complex
highway and traffic safety departments of many countries network performing their jobs based upon visual, hearing
shows the alarming figures of 1.4 million deaths and up and sensing etc. Therefore, an increase in the global level
to 50 million minor or serious injuries [4]. The Highway of coherence of cortical electric activity which indicates the
and Traffic Safety Administration (USA) has estimated that drowsiness [12]–[14].
around 100,000 car crashes are due to drowsiness, sleepy Drowsiness is associated with a gradual decrease in
feeling or fatigue [1], [3], [4]. Figure 1 plots the highway response time, less vigilant behavior, deficiency in processing
car crash data for the last 25 years in the USA. Sleep is a of available information and errors in short-term memory.
neuro-biological need of healthy humans [5]–[7]. In general, The driver’s drowsy state or sleepy behavior results in loose
sleepiness/drowsiness can be defined as slow and gradual loss control of his car, which later might collide with either
other moving vehicles or certain stationary objects [3], [10].
The associate editor coordinating the review of this manuscript and Several researchers and car manufacturers are working hard
approving it for publication was Chao Yang . to develop methodologies that can be used to prevent such

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
167172 VOLUME 7, 2019
M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 1. Car crashes site information in cases of driver’s

drowsiness/sleepiness.

the time delay between each heartbeat, ECG signals, res-

FIGURE 1. Total number of car accidents per year as observed by National piration rate, pulse pattern, etc., [26], [40]. In addition to
Highway Traffic Safety Administration (NHTSA), USA.
these measurements, nowadays, cortical signals are also an
indicator of different brain conditions, especially the drowsy
accidents [2]. There are two possible ways to avoid acci-
state [12], [17], [19], [24], [40]–[44].
dents. The first is the use of preventive measures, by which
Accidents that result due to the drowsy state of drivers
a driver is monitored, and an alarm is displayed to advise
have certain common features that traffic safety personnel
him/her to stop the car during a dangerous mental stage.
normally observe. These features include but are not limited
The second is the abnormal situation in which a driver is
to (1) driving during late-night hours (2:00 AM to 6:00 AM)
deemed to be asleep and car systems automatically control
or after mid-afternoon (3:00 PM to 5:00 PM); (2) driv-
the vehicle to avoid accidents until it can be safely stopped.
ing on highways/expressways; (3) drifting of only one car;
Cortical-activity-related signals are nothing but measures of
(4): young drivers; (5) non-application of brakes; (6) one per-
chemical ion change/signaling. These chemical ion changes
son only is in the car; (7) lack of safety measures at accident
can be measured through positron emission tomography
sites; (8) blood-alcohol level of the driver being over the limit
(PET), or by blood oxygen level dependent (BOLD) func-
for safe driving; (9) vehicle drifting from the road or colliding
tion via functional magnetic resonance imaging (fMRI)
with another vehicle [29], [45]–[49]. Table 1 summarizes the
[15]–[17], the hemodynamic-related signal detected by func-
symptoms observed of car crashes resulting from a driver’s
tional near-infrared spectroscopy (fNIRS) [18]–[20], and
drowsy state.
indirect measurement of electrical activity by electroen-
cephalography (EEG) [13], [21]. These cortical-activity- II. DROWSINESS AND ITS CLINICAL SYMPTOMS
measuring techniques were initially used to compare main A comfortable and good level of sleep is a crucial part of body
signals during activity and resting states and, thereby, to char- health. Sleep is defined as a specific time span reserved for
acterize externally applied stimuli [22], [23]. However, it is a the rest of the human body and cortical functions to reduce
well-established fact that substantial electric and metabolic body and mental fatigue for healthy functionality [8]. It is
activity exists in the brain even during the resting state of logical that during sleepiness/drowsiness, the consciousness
wakefulness [12]. of the brain is partially suspended [8], [29]. It is very difficult
Over the years, researchers have developed methodolo- to measure the sleep time required for healthy functionality of
gies to predict or indicate the drowsy state prior to a colli- the brain and human body and more specifically the specific
sion [8], [24], [25]. Generally, these methodologies can be time of each stage of sleep [12], [50]. The National Sleep
categorized into three main streams: vehicle-based sensors Foundation (NSF; USA) recommended that a new-born baby
that predict a drowsy or unusual state [26]–[29]; observa- needs 14-17 hours (h) of sleep, that infants require 12-15 h,
tion of driver-behavioral data [30], [31], and measurement toddlers 11-14 h, preschoolers 10-13 h, primary-school-going
of drivers’ physiological signals [32]–[35]. Vehicle-based children 9-11 h, teenagers 8-10 h, young adults 7-9 h, adults
sensing may be done by steering wheel data, steering wheel 7-9 h, old people 7-8h∗. A pictorial view of sleep require-
angle, the applied pressure pattern on an acceleration pad- ments is provided in Figure2.
dle, lane-position indication by an external visual sensor, Normally rapid eye-movement (REM) sleep, on average,
and/or a pressure sensor installed inside of the driver’s accounts for 20- 25% of total sleep time. Sleep depriva-
seat [36]–[38], among other means. The behavioral aspects of tion/drowsiness at any time of day is one of the symptoms
a driver may be determined by continuous recording through indicating an inadequate duration of sleep [8]. Such symp-
a camera installed in the dashboard to observe eye-closure toms are dominant during the late hours of the day and
time, eye-blinking frequency, movement and pose of the the early morning [29]. Alcohol and drugs worsen sleep
head, yawning and eye openings, etc., [36], [39]. Simi- deprivation/drowsiness symptoms, and caffeine/tea has less
larly, physiological data on a driver can be observed to effect in overcoming sleep deprivation than is commonly
detect the drowsy state. These signals include heart rate, supposed. The drowsiness feeling that occurs during daytime

VOLUME 7, 2019 167173

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 2. Symptoms, causes and reduction strategies for drowsy state.

is relaxing. The later stages of stage 1 produce further low-

frequency signals in the range of theta waves [53]. Stage 2:
In this stage, the slow movement of the eye balls starts to be
discontinuous; this is the first stage of NREM. It is not easy
to stay awake by low-intensity external stimulation. There are
two characterizing features during this stage, namely, sleep
spindles and k-complexes [54]. During stage 2, the body tends
to move into a deep/further-relaxation mode with dominating
theta wave activity including bursts of spikes in electrically
measured signals known as sleep spindles. Thus, sleep spin-
dles are high-frequency cortical-electrical signals that exist
in the form of low-frequency waves. Similarly, k-complex is
a very low-frequency pattern with relatively high amplitude
FIGURE 2. Sleep requirements by age group for healthy functionality of
daily life. signals, and is often measured in stage 2. Stages 3 &4:
These stages are referred as deep sleep with relatively low-
frequency waves less than 4 Hz with relatively high ampli-
is caused by several factors [26], [29]. The first and most
tudes [7], [8]. The frequency of waves in these stages are
important factor is poor sleep quality. Insomnia is one cause
those of delta waves [11]. A human is less effected by external
of poor sleep quality. It is characterized by a person’s unsuc-
stimulation during this stage. REM Sleep: This stage of
cessful efforts to sleep that last for more than one-half hour.
sleep is referred to as rapid eye movement sleep (REM). The
Also includes periodic disturbances during sleep. Another
cortical signals observed in this stage are very similar to those
fundamental reason for poor sleep quality is an uncomfort-
measured during wakefulness [55]. This is commonly known
able sleeping environment. It includes uncomfortable light,
as dreaming sleep, and all muscles are paralyzed with the
acoustic disturbances, etc. In addition, inadequate breathing
exception of those that are related to the respiratory system,
flow, with an insufficient amount of oxygen, can be a cause of
cardiopulmonary system, etc. Sleep deprivation or drowsi-
restless sleep that results in a drossiness state during daytime.
ness symptoms prevailing for longer times are associated with
Table 2 summarizes drowsiness symptoms and causes.
negative consequences for the human body and functionality.
Pressure on the eyes, fatigue and headache, and restless-leg
Table 3 below shows the systems and frequencies of cortical
syndrome are major indications of the drowsiness state.
signals, biological processes and physiological signals during
Drowsiness can be overcome by 15- 20 minutes of sleep
each stage of sleep.
(a nap), a cup of coffee, a shower, and other means.

III. SLEEP CLASSIFICATION AND STAGES IV. EVALUATION OF DROWSINESS/FATIGUE

The sleep/drowsiness process can be divided into four stages AND SLEEP LEVEL
of sleep [12], [51]. These stages are termed as stage 1, stage It is very important to analyze and evaluate the level of
2, stage 3 and stage 4. The first three stages are part of drowsiness. The literature shows that there exist several meth-
non-rapid eye movement (NREM), and the final stage is ods and strategies that can be utilized to measure drowsiness
named rapid eye movement (REM) in the literature [52]. levels. The details of these methods are given below.
Stage 1: During stage 1 the body feels fatigue and tends to
relax, and this may cause jerks and abrupt changes in body A. EPWORTH SLEEPINESS SCALE
position. The drowsiness level in stage 1 can easily be dis- Dr. Murray Johns (Melbourne, Australia) developed a method
connected/disrupted by external stimulation of low intensity. of assessing average sleep and drowsiness levels [56]. This
Mostly, stage 1 is related to the alfa range of an EEG signal. method consists of a simple questionnaire having eight ques-
The patterns and characteristics of the cortical signal during tions with numeric answers in the range of 0-3. A cumulative
stage 1 are similar to cortical activity related to someone who score of eight answers is divided among five groups: normal

167174 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 3. Sleep stages and corresponding biomedical information.

sleep (0- 5), higher normal sleep (6- 10), mild-excessive reduction of heart rate is also observed at nighttime while
sleep (11-12), moderate-excessive sleep (13- 15), and severe driving during late hours. The second procedure entails detec-
excessive sleep (16- 24) [57]. This is the most utilized method tion of heart rate variability, i.e., the time interval between
for analysis of drowsy/sleep stages, especially during day- pulses [73], [74]. The previous literature shows that the heart
time [56]. rate variability of an awake person manifests high-frequency
signals, whereas sleep-deprived or drowsy persons show low-
B. MULTIPLE SLEEP LATENCY TEST frequency heart rate variability [75]–[77]. Respiration rhythm
This test is based on the fact that persons who fall sleep is another important parameter that can be a potential dis-
quickly are considered to be drowsier [58]. This test mea- criminatory feature for the drowsy/alert states. According to
sures small intervals of sleep-oriented situations throughout a the literature, the frequency of the breathing pattern decreases
day [59], [60]. On the basis of sleep stages, it is concluded that during the drowsy state.
a person is drowsy, if he falls asleep within five minutes [61]. F. ELECTRO-OCULOGRAM (EOG)
The multiple sleep latency test (MSLT), which measures the This method estimates alertness level based on eye move-
ability to fall asleep under soporific conditions, is the current ment [78]. Specifically, it measures the electrical signal repre-
gold standard for quantifying daytime sleepiness [62]. senting the potential difference between the cornea and retina
of a subject’s eye [79], [80]. In this method, an electrode is
C. STANFORD SLEEPINESS SCALE placed near the end of each eye and another is placed at top
The Stanford sleep detection system consist of seven state- of the eye. Later, signals are classified for eye movement.
ments through which a person’s level of wakefulness is eval- In cases of a slower rate of eye movement than is normal,
uated [63], [64]. A subject chooses a single number from the subject is declared to be in the drowsy state [81]. Several
a list by evaluating personal symptoms mentioned in that studies in the past have utilized this methodology to access
list [65], [66]. the drowsy state with accuracy [78]–[82].
G. HEAD MOVEMENT
D. KAROLINSKA SLEEP TEST
This is a similar test to the Sandford sleepiness scale. A sub- It is a well-established fact that during the sleepy/drowsy
ject rates his/her current feeling by choosing a number state, the human body starts to feel relaxed due to fatigue
from a list of levels 1-9, depending upon the closest situ- caused by work load [83]. Thus, the head movements of
ation mentioned in each level [67]. The Karolinska Sleep normal and drowsy persons will have different characteris-
Test (KST) stages are divided into ‘‘very-alert’’, ‘‘very- tics [84]. A drowsy person will show head nodding, means
sleepy’’, ‘‘fighting-sleep’’, and ‘‘an effort to remain awake’’ to say, sudden changes of head position. This condition can
based on nine-steps [68]. It has already been validated against be measured using visual or motion sensors. This method has
physiology and performance measures [69]. been shown to be a potential choice for detection of drowsy
stages [83]–[87].
E. ELECTROCARDIOGRAM (ECG) H. JAW MOVEMENT
In this methodology, a human subject’s heart rate is evalu- It is common practice for a wakeful person to inhale air with
ated [70], [71]. There are two possible procedures. In the ordinary movement of the jaw [88],whereas a tired person
first one, heart rate is measured. It is a well-established will open up the jaw much more widely [89]. This can easily
fact that during sleep stages or drowsiness, the heart rate be measured through a simple camera that operates based on a
starts to decrease [72]. It is important to mention, too, that pattern recognition methodology. Several studies in past have

VOLUME 7, 2019 167175

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

experimented with human subjects and animals for detection

of sleep symptoms [88]–[92].

I. EYE BLINKING
Eye blinking is another discriminatory feature for detection
of alertness [93]. It can be observed in several ways. One of
them is to detect the opening of the eyelids. It is common
that during fatigue situations, the gap between eyelids tends
to decrease. Similarly, another way could be to measure eye-
blinking frequency. A drowsy-state person tends to manifest
a high eye-blinking frequency as compared with an awake
person [93]–[95].

V. NEUROLOGICAL EVIDENCE OF DROWSINESS

Technological development has enabled humans to measure
cortical signals indicative of different brain states [96]. These FIGURE 3. Publication record for drowsiness, sleepiness and fatigue
according to neuro-imaging modalities.
findings can be helpful for avoidance of car crashes and
prediction of brain states much earlier than the points at which
collisions happen [19], [29], [51]. fNIRS, EEG and fMRI,
being non-invasive methodologies, are useful for such inves-
tigations [37], [97]–[101]. Among them, EEG and fNIRS
are at competitive positions due to additional advantages of
portability, low cost, and acceptable temporal and spatial
resolutions [96], [102], [103]. In past, different sensors have
been utilized by car manufacturers to detect and indicate
drivers’ drowsiness state. Nonetheless there are still many car
crashes due to drowsy feelings. Therefore, several researchers
have initiated efforts to find and predict drowsy states using
neuro-activation data. EEG-based drowsiness/sleep-related
findings have been published most frequently in the past
50 years. Also, the symptoms of drowsiness and their corre-
sponding effects on the hemodynamic response have begun
to be more closely scrutinized. We have analyzed the pub-
lished data of drowsiness/sleep/fatigue studies and found very
interesting results. Figure 3 summarizes the details of pub- FIGURE 4. Percentages of published articles addressing drowsiness (D),
lished articles related to drowsiness/sleepiness and fatigue sleepiness (S) and fatigue (F) according to EEG (E), fMRI (M) and fNIRS (N).
based on EEG, fNIRS and fMRI. All the statistics presented
in Figure 3 is collected through Web of Science by putting also calculated driving-condition level scores to discrim-
different keywords. The keywords used for search are EEG + inate well-rested and sleep-deprived persons. Khan et al.
drowsiness, EEG + sleep, EEG + fatigue, fMRI + drowsi- (2015) [19] implemented linear discriminate analysis (LDA)
ness, fMRI + sleep, fMRI + fatigue, fNIRS + drowsiness, to detect drowsiness by utilizing eight different features from
fNIRS + sleep and fNIRS + fatigue, A relative-percentage optical data with different windows of 0∼15 seconds. Their
analysis of the published results also is presented in Figure 4. results suggested that mean accuracies of 83.1% to 84.9%
They show that most of the drowsiness-related studies pre- could be achieved by using these features: mean HbO, sig-
senting neuronal-activation-based findings have been based nal peak, and sum of peaks. The authors have divided the
on EEG. scanning area into three regions, concluded that, on aver-
age, dHbO has a relatively higher concentration during the
A. FUNCTIONAL NEAR-INFRARED SPECTROSCOPY drowsy state. Li et al. (2018) [116] evaluated 13 individ-
(FNIRS)-BASED FINDINGS uals ina seven-hour driving simulation test by observing
Ahn et al. (2016) [40] evaluated the neuro-physiological cor- optical signals from the pre-frontal cortex. They found that
relates of drivers’ mental conditions. This study compared a specific area therein showed increased dHbO and dtHb
well-rested individuals with sleep-deprived persons using as the time of driving increased. Nugyun et al. [5], [97]
eight-channel fNIRS in addition to EEG and biological sig- analyzed fNIRS and EEG signals in addition to biological
nals. The authors found that the dHbO of the well-rested signals to discriminate drowsy and awake states. They uti-
controls was higher than that of the sleep deprived. The dHbR lized classical LDA and found that dHbO and the beta power
of both groups showed much closer patterns. The authors in the frontal lobe were significantly different during the

167176 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

two states. They found that the mean values of the awake hemodynamic signals under visual fatigue. Their findings
and drowsy states for dHbO were 0.019 and −0.017, respec- suggested that hemodynamic activity during visual fatigue
tively, and for dHbR, were −0.003 and 0.005, respectively. is influenced by time. Xu et al. (2017) [42] analyzed the
Borragen et al. (2018) [117] analyzed mental fatigue using functional connectivity for a long driving simulation task
trans-cranial current stimulation (tDCS) and observed cor- coupled with mental calculation tasks in order to investigate
tical oxygenation changes by fNIRS in the frontal cortex. fatigue. They concluded that the pre-frontal and motor cortex
They concluded that tDCS, in comparison with the sham con- are closely connected during mental fatigue. Harrivel et al.
dition, failed to counteract cortical-fatigue-related changes, (2013) [18] analyzed functional neuro-imaging data to dis-
whereas tDCS combined with induction of cortical fatigue criminate alertness and rest periods. The authors defined two
shifted the inter-hemispheric oxygen level in the post-training regions: one related to high levels of task engagement, called
resting state. In an interesting study by Liu et al. (2014) the task-positive region, and the other related to low levels
[110], a positive correlation between drowsy state and pre- of engagement, called the task-negative region. The authors
frontal activation was found by experimenting with a simu- claimed that for the first time, it was possible to detect,
lated environment in which a speed-control driving task was through fNIRS, a negative correlation between HR activity in
performed. The level of drowsiness was evaluated through the key areas of the task-positive and task-negative regions.
a five-item Likert-type questionnaire. An increase in pre- These methods could be helpful for monitoring of atten-
frontal activation was observed via fNIRS during the drowsy tion/alert states using fNIRS. Bu et al. (2018) [137] effec-
state. Cai et al. (2017) [118] investigated fatigue level using tively showed that wavelet phase coherence (WPC) and the
fNIRS by parietal-occipital scanning. They found a spatial wavelet amplitude of dHbO are discriminatory features for
correlation between the occipital cortex and binocular depth analysis of poor-quality sleep. They analyzed 15 poor-quality
perception. Chuang et al. (2018) [119] investigated four sleep (PQS) individuals in comparison with 14 healthy con-
different regions, namely the parieto-occipital, motor-right, trols, having scanned their pre-frontal, sensory motor and
motor-left and frontal cortex, to discriminate the drowsy occipital lobes using fNIRS. They found that the WPC values
state. They concluded that there was a significant increase of PQS were lower than those of the healthy individuals. This
in dHbO power in those regions during the deviation periods. finding suggests that individuals could be scanned before
In an interesting work by Karageorghis et al. (2018) [134], long driving to avoid any drowsiness condition in cases of
the authors found that music behaves as a potential candidate poor sleep. Chaung et al. (2018) [119] tested16 individu-
for prevention of internal fatigue-related hemodynamic sig- als in an event-related lane-departure experiment of 1 hour
nal measured through fNIRS. Mehta and Rhee (2017) [135] duration. The results indicated that dHbO increased while
analyzed social stress and its effects on pre-frontal cortex the driver was fighting fatigue. Table 4 summarizes the work
HR signals by experimenting with 60 adults and children done to detect drowsiness by utilizing fNIRS and the corre-
based on motor tasks. The results suggested that aging results sponding contributions.
in decreased force-steadiness with a significant increase in
bilateral pre-frontal cortex activity. In an earlier work, Mehta B. FUNCTIONAL MAGNETIC RESONANCE IMAGING
et al. (2015) [136] concluded that stress fatigue has different (FMRI)-BASED FINDINGS
patterns for fat/obese people than for people of normal BMI. The characterizing feature for any cortical imaging devices
It could therefore be concluded that obese persons must be are temporal and spatial resolutions [96]. Functional mag-
more careful while driving because they may have drowsy netic resonance imaging (fMRI) has better spatial resolu-
feeling much quicker than non-obese people. A very impor- tion than EEG or fNIRS [104]. However, at the same time,
tant study by Muthalib et al. (2013) [120] investigated mental it entails a huge setup, making it unsuitable for any portable
fatigue in relation tothe motor- and pre-frontal cortices by job [105]. Additionally, regular analysis of a driver’s drowsi-
stimulation via trans-cranial direct-current stimulation. The ness state by fMRI is very difficult. On the other hand,
pre-frontal cortex was scanned through fNIRS while for the the findings of fMRI can be beneficial to any brain-state anal-
motor cortex, tDCS was applied. They found higher activity ysis seeking to localize specific brain areas showing response
in the pre-frontal cortex during task failure irrespective of the under sleepy/drowsy and alert/awake conditions [16]. Also,
tDCS-application conditions. Ochi et al. (2018) [121] pre- fMRI data’s correlation with EEG and fNIRS data offers
sented results for pre-frontal activation in a hypoxic environ- a great potential for clearer and more concrete findings
ment based on an analysis of15 healthy subjects performing for those two latter modalities [12], [106]. In this context,
a color word strop task both before and after a 10-minute Allen et al. (2018) [125] analyzed the connectivity patterns
moderate-intensity exercise. They concluded that hypoxic of brain networks during eyes-open and eyes-closed states.
exercise reduces left-dorso-lateral pre-frontal cortex activity. Several existing methodologies have been used to explore
Such conditions could be related to the enclosed environment functional connectivity and its usefulness for understand-
of a car during a long drive. Rhee and Mehta (2018) [122] pre- ing and characterizing brain states responsible for partic-
sented results showing that females have more motor fatigue ular tasks [107]. Allen et al. (2018) [125] suggested that
than males based on an analysis of obese and non-obese per- it is very important to analyze drowsiness level in both
sons. Richter et al. (2018) [123] analyzed pre-frontal cortex eyes-closed and eyes-open states, and indeed, the authors

VOLUME 7, 2019 167177

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 4. Summary of fNIRS-based neurological findings related to drowsiness, fatigue and sleep stages.

showed that –the two states respond in common and also a divided-attention task utilizing fMRI. Ong et al., (2015)
manifest different connectivity patterns from those associated [127] analyzed spontaneous eye-closure activity. Such activ-
with EEG-spectral signatures. In another study, Tagliaznc- ity is frequently adopted by individuals insleep-deprived
chi and Laufs [108] and Tagliazncchi et al. [109] analyzed conditions. The authors observed concurrent and extensive
functional connectivity under the sleep and awake conditions. hypnagogic co-activation of the visual, auditory and somato-
Their findings indicated an increase in network modality sensory cortices and DMN as well. Since spontaneous fluc-
during the N3 sleep stage in comparison with the alert/awake tuations of the eye are indicative of the drowsy state, visual
state. Haimovici analyzed brain network connectivity pat- auditory and DMN could be scanned through portable devices
terns during wakefulness. They observed that the clustering for determination of sleep-deprived drivers and fatal acci-
of dynamic functional connectivity upon short time windows dents could be avoided. Poudel et al. (2014) experimented
reflected fluctuation in wakefulness. This report validated with 20 healthy humans in the well-rested condition (i.e.,
the previous findings of increased modularity during non- they were not sleep deprived). The participants were put
REM sleep. Horavitz et al. [12] investigated different lev- through a 50-minute driving task. All of the experiments
els of consciousness (i.e., alert state)under light-sleep and were recorded through a camera. The visual recordings
drowsy conditions. They found a significant increase in showed micro-sleep behaviors of around 70% of the sub-
BOLD-signal fluctuations in the visual cortex in addition jects. The fMRI recording during the experiments suggested
to other cortical areas, though the visual cortex fluctuations increases/decreases of activity patterns in different brain
were the most significant. For the drowsy state, meanwhile, regions. Thalamic posterior-cingulate and occipital cortex
they observed a correlation with the default mode network activities were observed to be decreased, whereas frontal,
(DMN). Their findings suggests that DMN and primary posterior and parietal activities were found to be increased,
sensory cortices activities do not require a level of conscious- during short intervals of sleep recorded through the camera.
ness. Drummond et al. [55] observed increased pre-frontal Stern et al. (2011) [6] analyzed the source localization and
activation during the drowsy state by experimenting with function of vertex sharp transient, which is a marker of the

167178 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 5. fMRI-based previous studies for detection of drowsiness/ sleep stages.

NREM sleep stage in EEG data. The simultaneous recording EEG activity from 11 subjects in a 40-45 kph driving experi-
of EEG and fMRI showed that the primary sensory cortex is ment. The results of their analysis showed that periodic speed
responsible for sharp transient ejection. The drowsy feeling modifications can be a way to stay alert, whereas driving
could also be the result of motor fatigue. Duinen et al. (2007) at a constant speed, which tends to bore drivers to certain
[129] comprehensively analyzed the relation of fatigue with degrees, may cause fatigue and drowsiness. Eoh et al. [11]
cortical activation. Their observation showed an increase in studied the drowsiness/fatigue in eight subjects in a simulated
motor and bilateral orbito-frontal activation with increasing driving experiment. Fifty minutes of 8-channel EEG data and
motor activity in cases of fatigue. However, the activation subject behavioral data obtained by camera was recorded and
clusters were small. The brain stem is involved in the control analyzed. Three frequency bands, namely theta, alpha and
of several important brain functions that operate the human beta, were utilized in terms of their mean power, ratio indices
body. These functions include respiration rhythms, heartbeat, and burst activity to analyze the sleepy states of the subjects.
and sleep cycles, among others. It is difficult to record func- The results showed that the alpha, beta, index beta/alpha
tional data of the brain stem through fMRI, since its location and index (alpha+ theta/beta) were significantly different
is very close to major arteries adjacent to pulsatile cere- as the drivers’ state was converted from alert to drowsy.
brospinal fluid-filled spaces [41]. Thus, alternatively, indirect Pal et al., 2008 [51] proposed an unsupervised method to
measurement or some network activation can be analyzed detect changes in the alert state of drivers. They suggested that
through fMRI during sleep or drowsy feelings. Sleepiness in the alert state, PSD from the theta and alpha bands at chan-
behavior and cortical responses vary over the course of the nel Oz follows a multivariate normal distribution, which can
24 hours of each day, depending on sleep conditions, work be used to drive alert model of each subject by using a small
load, and, specifically, cardiac pulsations. Maire et al. (2018) amount of EEG data. Then, this model was used to analyze
[130] experimented with healthy individuals by performing a the deviation in PSDs to determine whether the driver is in
psychomotor vigilance task in fMRI under two conditions. the alert state or the drowsy state. In another study, Yeo et al.,
The first one is sleep deprivation based on40 hours of no 2009 [13], the authors used the SVM classifier to detect the
sleep, which group was deemed the high sleep pressure group. alert and drowsy states of drivers. Twenty young participants
The second group, deemed the low sleep pressure group, was were participated in a simulated driving task to acquire EEG
allowed 40 hours of a multiple nap-protocol. The sleepiness signals. Features were extracted by calculating the PSD of the
level was evaluated on the Karolinska sleepiness scale. The delta, theta, alpha and beta bands. For each frequency band,
experimental data suggested that cardiac and sleep homeo- four features, namely dominant frequency, average power of
static processes impact upon vigilant attention. Table 5 sum- dominant peak, center of gravity frequency, and frequency
marizes the work done to detect drowsiness by utilizing variability, were calculated using PSD values. Finally, SVM
fMRI, along with the corresponding contributions. was used to classify the alert and drowsy states of drivers.
The results suggested that SVM can achieve a classification
C. ELECTROENCEPHALOGRAPH (EEG)-BASED FINDINGS accuracy of 99.3%, indicating that it can be successfully used
In 2002, Tejero and Chóliz [9] studied should the effects of to classify alert and drowsy states. Simon et al., 2011 [52]
constant speed and periodically varying speed on the alertness proposed an algorithm that uses a few seconds of EEG data
of drivers in a motorway driving experiment. They recorded to detect alpha spindles (short narrowband bursts in the alpha

VOLUME 7, 2019 167179

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

band) by measuring peak frequency, amplitude, and duration. Hajinoroozi et al., 2016 [50], proposed a channel-wise con-
They tested their algorithm on signals generated both synthet- volution neural network with a restricted Boltzmann machine
ically and acquired under real-life driving conditions. They algorithm for classification of drivers’ different states. EEG
then compared the results of their algorithm with the con- data from 37 subjects acquired during a virtual-reality-
ventional alpha power criterion. Statistical analysis revealed based driving simulator experiment were used in this study.
significant increases from the first to the last driving section The results showed that the proposed algorithm achieved
for the alpha spindle parameters. Another study, Liu et al., 76.72% cross-subject accuracy and outperformed other algo-
2013 [111] used a single-channel wireless EEG system to rithms such as SVM and DNN. Yang et al. 2018 [54], pro-
detect the drowsy state of drivers by incorporation of music to posed a two-layer algorithm by analyzing the behavioral
reduce the chances of becoming drowsy. Data acquired from data and EEG signals acquired from 52 participants in a
20 male and 20 female subjects were decomposed into four simulated driving experiment. In the first layer, 14 fea-
different frequency bands to calculate PSD features for the tures were extracted from the behavioral data acquired from
classifiers. An integrated classifier consisting of SVM, ANN the simulation software. Five categories of driving behavior
and kNN was used to achieve an average accuracy of 81.3% were obtained based on the driving data and the following
in detection of the drowsy state. In a later study, Gurudath procedure: first, K-means was used to form two clusters
et al., 2014 [112], the authors developed a drowsiness detec- of base driving features, respectively the driving style and
tion algorithm by combining discrete wavelet transform and stability. Then, SVM and SVM recursive feature elimination
K-mean clustering. EEG data acquired from a driving sim- were applied for feature selection and extraction. Finally,
ulator experiment was decomposed into different frequency the driving-behavior types were classified using K-means.
bands using discrete wavelet transform. The mean, median, In layer two, EEG features extracted from different frequency
variance, standard deviation and mode were calculated for bands were classified using the kNN classifier. The results
each band and provided as features to the K-mean clustering suggested that their methodology can achieve an average
algorithm for classification. Roy et al. (2014) [21] proposed classification accuracy of 69.5%, and a maximum accuracy
an ICA and adaptive thresholding-based algorithm to detect of 83.5%. Chen et al., 2018 [132] proposed a novel brain-
eye blinks using EEG data only, whereas most of the compa- networks-based approach to detect drowsiness. EEG sig-
rable studies have used EOG signals to do so. They suggested nals from 15 male subjects were acquired in a simulated
that this algorithm can be used to detect the mental fatigue of driving task. They calculated phase lag index (PLI) values
drivers. They acquired data from 11 subjects for a working for different frequency bands to assess the functional con-
memory task. The results of their analysis suggest that their nectivity between different pairs of channels and to con-
algorithm can achieve a true detection rate of 89%, while the struct a minimum spanning tree by connecting electrodes
correlation of their results with the results from EOG signals in descending values. The results of their statistical anal-
was 0.81. Li and Chung (2014) [14] hypothesized that there ysis showed that the PLI values in the delta band tend to
is a linear relationship between EEG activity in the occipital increase as the subject’s state transitions from altered to
cortex (specifically the O2 location) and eye-closure degree. drowsy. In their subsequent work [113], the alert and drowsy
They analyzed EEG data from 30 subjects, and their results states of drivers were analyzed by a proposed combined min-
suggested that alpha power at O2 increases with increased imum spanning tree and synchronization likelihood-based
drowsiness, whereas beta power decreases with increased functional brain network approach. Features extracted from
drowsiness. Their proposed model achieved 87.5 and 70.0% these were transferred to four well-known classifiers, from
accuracy for the male and female subjects, respectively. In a which the K-Nearest-Neighbors classifier achieved a max-
subsequent work, Li and Chung (2015) [53], developed a imum average accuracy of 98.6%. Recently, Wang et al.,
Bluetooth-based low-power-consumption system to detect 2018 [114] used dry EEG to evaluate two methodologies
early drowsiness of drivers. They combined EEG data with based on PSD and sample entropy to detect drivers’ fatigue
head movement data to increase the SVM-based alert/drowsy in real-life driving. To do so, they acquired EEG signals
classification accuracies. Data from six subjects who had from 10 subjects in a simulated driving experiment. The
participated in a simulated driving exercise yielded 96.24% results of their analysis suggested that PSD scores increase
accuracy when EEG data is used with head movement, with increased mental fatigue, whereas sample entropy
as compared with 82.71% when only EEG data was used. scores decrease with increased mental fatigue. However, they
Gharagozlou et al., 2015, [131], analyzed the fatigue caused observed that PSD is more robust and more effective than
to drivers using a virtual reality simulator. Twelve healthy sample entropy to detect mental fatigue while driving. In
male car drivers participated in an overnight study. They another study, Wang et al., 2018 [44], the authors developed a
analyzed the initial and final 10 minutes of recordings single-channel-EEG-based method to detect driver fatigue
by calculating the PSD and FFT in the alpha frequency by combining four features, three base classifiers and three
band. The results showed that there was a significant dif- ensemble algorithms. They tested EEG data acquired from
ference in the absolute alpha power between the initial and 12 subjects in a simulated driving task for different combi-
final stages; moreover, the increase in the alpha power in nations of features, classifiers and ensemble algorithms. The
the final section of driving indicated the onset of fatigue. results of their analysis suggested that Channel T6, feature

167180 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

fuzzy entropy, and classifier-gradient-boosted decision tree the pattern of driving and, if necessary, declares the driver
with 500 boosting stages achieved the best results, boasting an un-attentive by displaying a coffee cup signal and sounding
average accuracy of 91.1%. Zhang et al., 2018 [115] studied an audio alert.
driver drowsiness using EEG to find the optimal recording Mercedes’ ‘‘Attention Assist System’’ determines the
locations. Twenty-two subjects were used for recording of driver’s condition through a specific sensor installed in
EEG signals from the Fp1, Fp2, T3 and T4 locations in a the steering wheel in addition to observing the acceleration
simulated driving experiment. The PSD was calculated for the of the car, breaking patterns, the condition of the road and
delta, theta, alpha and beta frequency bands. The results sug- driving time, etc. In contrast with other cars, which utilize
gested that delta band activity increases, whereas theta and visual sensors to monitor facial expressions and specifically
alpha activity decreases in the drowsy state as compared to the the eye patterns representative of the driver’s current state,
alert state. They further suggested that changes at the T3 and Mercedes has installed its specific steering sensor. The over-
T4 locations were much larger than changes at Fp1 and Fp2. all system continuously observes the driver’s driving patterns
Barua et al. 2019 [24] studied sleepiness during driving using during a particular trip and generates a driver’s profile with
EEG and EOG. They recruited 30 drivers to drive in the certain thresholds. If the system detects a significant variation
alert and sleepiness states within a simulated driving scenario. in comparison with the generated profile, the Attention Assist
Three conditions, namely alert, somewhat sleepy, and sleepy, System declares the driver in the fatigue stage and gener-
were analyzed. EEG features were extracted by calculating ates an alert or warning sign and beep. Ford has installed a
PSD for the different frequency bands, whereas blink duration ‘‘Driver Alert System’’ that is based upon lane assist data.
was used as the feature from EOG. Four different classifiers A pinhole camera installed behind the dash rearview mirror
were used to classify the multistate and binary state (alert continuously visualizes the moving patterns of the car and
vs. sleepy) conditions, from which SVM achieved maximum decides whether the vehicle is staying inside a lane. In cases
accuracies of 79 and 93%, respectively. Table 6 summarizes where the driver is deviating from a specific lane, the system
the work done to detect drowsiness by EEG and the corre- generates a vibration in the steering wheel to alert the driver
sponding contributions. and declare him to be in a drowsy state; and if the driver
does not respond within predefined time, the system adjusts
VI. AUTOMOBILE-INSTALLED DROWSINESS DETECTION the steering wheel to avoid deviation. Bosch has developed
SYSTEMS a ‘‘Driver’s Drowsiness detection system’’ that determines
The previously available data on road accidents indicated that the drowsy state of a driver by analyzing steering move-
around 25% of road accidents on highways/expressways are ment, and, if necessary, alerts him to apply the brake paddle.
caused by drowsy feeling, fatigue symptoms, etc. This statis- This is easily done by analyzing the car’s steering angle
tic showed that drowsiness causes more accidents even than and the velocity profile of that angle. Nissan has installed
drunk driving. Therefore, car manufacturers are working hard a ‘‘Drowsiness-Detective Alert’’ system that determines the
and financially supporting research to develop an intelligent drowsy state of a driver by analyzing the steering profile for
system installed in automobiles that not only alerts drivers a few minutes of driving. This system makes a model of
under drowsy conditions but also responds automatically and the driver’s behavior based upon steering wheel movement
intelligently to avoid accidents and collisions. Toyota has patterns (it is common practice for drivers to slightly adjust
developed a specific system to respond in cases of driver inat- and correct the steering wheel while driving under specific
tentiveness, namely, ‘‘Toyota Safety Sense’’. This system has road conditions). If the driver is in a drowsy state, he will
two main features. The first one is a pre-collision detection deviate from the model/observed data pattern and stop fol-
system that guards against frontal collisions by indicating lowing those particular characteristics, and/or he will slow the
visual and audio alarms to drivers, and, in cases where the pace of the pattern. In such situations, the dashboard of the
driver does not respond within a pre-defined time period, car displays a coffee cup notification to alert the driver and to
the system declares him to be in a drowsy state and automati- avoid any fatal accident. BMW has installed a ‘‘Driver Assis-
cally applies breaks to avoid collision. The second feature of tance System’’ that provides lane assistance, lane-departure
this system is correct lane alertness. In this hardware, a front- warnings and lane-change warnings based upon visual data
installed camera continuously detects the lane markers, and gathered through specific cameras installed at certain loca-
if the car deviates from those lanes, audio and visual beepers tions in the car. Table 7 summarizes the sensors installed
are displayed in the dashboard to correct the steering wheel by car manufacturers to detect drowsiness state and their
position. In cases where the driver does not respond to these working principles.
indications, the system declares the driver to be in a drowsy
state and automatically adjusts the steering. Hyundai has VII. RESEARCH CHALLENGES AND FUTURE
installed a ‘‘Driver Attention Warning’’ system. This system RECOMMENDATIONS
is based on the signal received from the wheel’s position in Drowsiness/sleepiness is a vital state of the human body
a particular lane and time of driving. By analyzing the angle that is essentially required for healthy functionality. How-
of torque applied to the steering wheel, the movement of the ever, in some cases, among which, obviously, is car driving,
car in the lane and the time of driving, the system determines drowsiness/sleepiness leads to devastating results. A recently

VOLUME 7, 2019 167181

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 6. Drowsiness/sleep-related findings in electrical signals of neuronal activity.

published article [133] have summarized some of methods also be the results of some other clinical symptoms of certain
frequently used for drowsiness detection and possible appli- diseases. Additionally, sun- and eyeglasses could also be
cations with certain classifiers. In this article we have tar- restricting factors by which eye-based drowsiness detection
geted brain signals and their respective response changes systems could fail. Skin temperature and heart rate might
during drowsy feeling in addition to separately analyzing result from the temperature inside the car cabin. In such
each method and corresponding pros and cons. Researchers conditions, the cortical-measured signal indicating drowsy
have endeavored to alert drivers during the drowsy state state is a potential candidate. However, equipment like fMRI
so that unwanted and devastating circumstances may is not suitable, since it requires a huge setup and is not
be avoided. At the same time, car manufacturers have been easy to apply for an experiment. On the other hand, it is of
exerting prodigious effort to develop drowsiness detection great potential importance, as some very important insights
systems and accident prevention mechanisms that not only into brain functionality can be gleaned from this device’s
accurately detect the drowsy state of a driver but also data. As for fNIRS,its portability and economy make it a
take automatic preventive action to avoid accidents. Still, promising candidate for drowsiness detection. Its drawback
though, there is no such system available that offers high is the un-comfortable feeling of the NIRS optodes positioned
accuracy. For example, lane-detection systems have some continuously over the pre-frontal cortex. EEG on the other
limitations: lane deviation is possible in overtaking and to hand, with its high temporal resolution, is the most frequently
avoid objects/obstacles. Similarly, installed steering sensors used biomedical system for drowsiness detection. Its specific
depend upon the accuracy of classification/detection algo- data features, k-complexes and sleep spindles are concrete
rithms, which cannot be 100%. In addition, false detection evidence of drowsy feelings. However, continuous installa-
and deliberate changes in steering profiles, due to some other tion of EEG electrodes on a driver’s head is a difficult task.
factors, are also possible. For example, eye blinking, eye- The present authors believe that a drowsiness detection
opening time, head nodding, and jaw movement, etc., could system could be designed by utilizing a multi-modality

167182 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

TABLE 7. Sensors installed by car manufacturers to detect drowsy [7] G. Taga, H. Watanabe, and F. Homae, ‘‘Developmental changes in cortical
feelings. sensory processing during wakefulness and sleep,’’ NeuroImage, vol. 178,
pp. 519–530, Sep. 2018.
[8] J. D. Slater, ‘‘A definition of drowsiness: One purpose for sleep?’’ Med.
Hypotheses, vol. 71, no. 5, pp. 641–644, Nov. 2008.
[9] P. Tejero and M. Chóliz, ‘‘Driving on the motorway: The effect of alter-
nating speed on driver’s activation level and mental effort,’’ Ergonomics,
vol. 45, no. 9, pp. 605–618, Jul. 2002.
[10] I. Filip, M. Tidman, N. Saheba, H. Bennett, B. Wick, N. Rousem,
D. Patriche, and A. Radfar, ‘‘Public health burden of sleep disorders:
Underreported problem,’’ J. Public Health, vol. 25, no. 3, pp. 243–248,
Jun. 2017.
[11] H. J. Eoh, M. K. Chung, and S.-H. Kim, ‘‘Electroencephalographic study
of drowsiness in simulated driving with sleep deprivation,’’ Int. J. Ind.
Ergonom., vol. 35, no. 4, pp. 307–320, 2005.
[12] S. G. Horovitz, M. Fukunaga, J. A. de Zwart, P. van Gelderen,
S. C. Fulton, T. J. Balkin, and J. H. Duyn, ‘‘Low frequency BOLD
fluctuations during resting wakefulness and light sleep: A simultaneous
EEG-fMRI study,’’ Hum. Brain Mapping, vol. 29, no. 6, pp. 671–682,
approach. For example, a steering sensor plus a dashboard Jun. 2008.
camera could be employed for continuous monitoring of [13] V. M. Y. Mervyn, X. Li, K. Shen, and E. P. V. Wilder-Smith, ‘‘Can SVM
be used for automatic EEG detection of drowsiness during car driving?’’
eye-blinking patterns, head nodding, and yawning, etc., Saf. Sci., vol. 47, no. 1, pp. 115–124, Jan. 2009.
in addition to a driving seat pressure sensor. A system with [14] G. Li and W.-Y. Chung, ‘‘Estimation of eye closure degree using EEG
multi-sensor data could result in much closer and accurate sensors and its application in driver drowsiness detection,’’ Sensors,
vol. 14, no. 9, pp. 17491–17515, Sep. 2014.
estimation of drowsy/sleep feelings. In addition to this, high- [15] J. Z. Liu, Z. Y. Shan, L. D. Zhang, V. Sahgal, R. W. Brown, and G. H. Yue,
way police can utilize an fNIRS/EEG system to monitor ‘‘Human brain activation during sustained and intermittent submaximal
driver at specific times. Additionally, as a simpler and easy fatigue muscle contractions: An fMRI study,’’ J. Neurophysiol., vol. 90,
no. 1, pp. 300–312, Jul. 2003.
option, self-evaluations in the form of drowsiness detection [16] J. C. de Munck, S. I. Gonçalves, T. J. C. Faes, J. P. A. Kuijer,
questionnaires could also be filled out by drivers at certain P. J. W. Pouwels, R. M. Heethaar, and F. H. L. da Silva, ‘‘A study of the
check points. brain’s resting state based on alpha band power, heart rate and fMRI,’’
NeuroImage, vol. 42, no. 1, pp. 112–121, Aug. 2008.
[17] K. Liu, G. Sun, B. Li, Q. Jiang, X. Yang, M. Li, L. Li, S. Qian, L. Zhao,
VIII. CONCLUSION Z. Zhou, K. M. von Deneen, and Y. Liu, ‘‘The impact of passive hyper-
This paper reviews drowsiness causes, systems, as well as thermia on human attention networks: An fMRI study,’’ Behavioural
Brain Res., vol. 243, pp. 220–230, Apr. 2013.
its preventive methodologies and neuronal activation states. [18] A. R. Harrivel, D. H. Weissman, D. C. Noll, and S. J. Peltier, ‘‘Monitoring
It is concluded that most of research work in brain signal attentional state with fNIRS,’’ Frontiers Hum. Neurosci., vol. 7, p. 861,
based drowsiness detection is done though EEG. Addition- Dec. 2013.
[19] M. J. Khan and K.-S. Hong, ‘‘Passive BCI based on drowsiness detection:
ally, no drowsiness detection and alert systems of cars are An fNIRS study,’’ Biomed. Opt. Express, vol. 6, no. 10, pp. 4063–4078,
hundred percent reliable. Therefore, multimodality approach Oct. 2015.
could be helpful and much accurate. [20] A. R. Harrivel, D. H. Weissman, D. C. Noll, T. Huppert, and S. J. Peltier,
‘‘Dynamic filtering improves attentional state prediction with fNIRS,’’
Biomed. Opt. Express, vol. 7, no. 3, pp. 979–1002, Mar. 2016.
CONFLICT OF INTEREST [21] R. N. Roy, S. Charbonnier, and S. Bonnet, ‘‘Eye blink characterization
The authors declare no conflict of interest. from frontal EEG electrodes using source separation and pattern recogni-
tion algorithms,’’ Biomed. Signal Process. Control, vol. 14, pp. 256–264,
Nov. 2014.
REFERENCES [22] J. Binder, J. Frost, T. Hammeke, P. Bellgowan, S. Rao, and R. Cox,
[1] A. Joly, R. Zheng, T. Kaizuka, and K. Nakano, ‘‘Effect of drowsiness ‘‘Conceptual processing during the conscious resting state: A functional
on mechanical arm admittance and driving performances,’’ IET Intell. MRI study,’’ J. Cognit. Neurosci., vol. 11, no. 1, pp. 80–93, Jan. 1999.
Transp. Syst., vol. 12, no. 3, pp. 220–226, Apr. 2018. [23] S. I. Rapoport, ‘‘Functional brain imaging in the resting state and during
[2] M. L. Macy, P. M. Carter, R. Bingham, R. M. Cunningham, and activation in Alzheimer’s disease: Implications for disease mechanisms
G. L. Freed, ‘‘Potential distractions and unsafe driving behaviors among involving oxidative phosphorylation,’’ in Oxidative/Energy Metabolism
drivers of 1- to 12-year-old children,’’ Academic Pediatrics, vol. 14, no. 3, in Neurodegenerative Disorders, vol. 893, J. P. Blass and F. H. McDowell,
pp. 279–286, May/Jun. 2014. Eds. New York, NY, USA: Annals of the New York Academy of Sciences,
[3] C.-T. Lin, C.-J. Chang, B.-S. Lin, S.-H. Hung, C.-F. Chao, and I.-J. Wang, 1999, pp. 138–153.
‘‘A real-time wireless brain–computer interface system for drowsiness [24] S. Barua, M. U. Ahmed, C. Ahlström, and S. Begum, ‘‘Automatic
detection,’’ IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 4, pp. 214–222, driver sleepiness detection using EEG, EOG and contextual information,’’
Aug. 2010. Expert Syst. Appl., vol. 115, pp. 121–135, Jan. 2019.
[4] Y. Choi, S. I. Han, S.-H. Kong, and H. Ko, ‘‘Driver status monitoring [25] M. S. Wang, N. T. Jeong, K. S. Kim, S. B. Choi, S. M. Yang, S. H. You,
systems for smart vehicles using physiological sensors: A safety enhance- J. H. Lee, and M. W. Suh, ‘‘Drowsy behavior detection based on driving
ment system from automobile manufacturers,’’ IEEE Signal Process. information,’’ Int. J. Automot. Technol., vol. 17, no. 1, pp. 165–173,
Mag., vol. 33, no. 6, pp. 22–34, Nov. 2016. Feb. 2016.
[5] T. Nguyen, O. Babawale, T. Kim, H. J. Jo, H. L. Liu, and J. G. Kim, [26] A. Chowdhury, R. Shankaran, M. Kavakli, and M. M. Haque, ‘‘Sensor
‘‘Exploring brain functional connectivity in rest and sleep states: A fNIRS applications and physiological features in drivers’ drowsiness detection:
study,’’ Sci. Rep., vol. 8, Nov. 2018, Art. no. 16144. A review,’’ IEEE Sensors J., vol. 18, no. 8, pp. 3055–3067, Apr. 2018.
[6] J. M. Stern, M. Caporro, Z. Haneef, H. J. Yeh, C. Buttinelli, [27] P. T. Gimeno, G. P. Cerezuela, and M. C. Montañés, ‘‘On the concept and
A. Lenartowicz, J. A. Mumford, J. Parvizi, and R. A. Poldrack, ‘‘Func- measurement of driver drowsiness, fatigue and inattention: Implications
tional imaging of sleep vertex sharp transients,’’ Clin. Neurophysiol., for countermeasures,’’ Int. J. Vehicle Des., vol. 42, nos. 1–2, pp. 67–86,
vol. 122, no. 7, pp. 1382–1386, Jul. 2011. 2006.

VOLUME 7, 2019 167183

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

[28] V. R. Kumar and S. Palaniappan, ‘‘Observation of drowsiness detection [50] M. Hajinoroozi, Z. Mao, T.-P. Jung, C.-T. Lin, and Y. Huang, ‘‘EEG-
based on vehicle measurement’s,’’ Res. J. Pharmaceutical Biol. Chem. based prediction of driver’s cognitive performance by deep convolutional
Sci., vol. 7, pp. 190–197, Jul./Aug. 2016. neural network,’’ Signal Process., Image Commun., vol. 47, pp. 549–555,
[29] A. Sahayadhas, K. Sundaraj, and M. Murugappan, ‘‘Detecting driver May 2016.
drowsiness based on sensors: A review,’’ Sensors, vol. 12, no. 12, [51] N. R. Pal, C.-Y. Chuang, L.-W. Ko, C.-F. Chao, T.-P. Jung, S.-F. Liang,
pp. 16937–16953, Dec. 2012. and C.-T. Lin, ‘‘EEG-based subject-and session-independent drowsiness
[30] M. Edraki, S. Zendehzaban, N. Beheshtipour, F. Hemmati, and detection: An unsupervised approach,’’ Eurasip J. Adv. Signal Process.,
S. Haghpanah, ‘‘Comparison of the effects of attachment training for vol. 2008, p. 192, Jan. 2008.
mothers on the behavioral responses of premature infants: A randomized [52] M. Simon, E. A. Schmidt, W. E. Kincses, M. Fritzsche, A. Bruns,
clinical trial,’’ Iranian J. Neonatol., vol. 6, no. 2, pp. 37–42, Jun. 2015. C. Aufmuth, M. Bogdan, W. Rosenstiel, and M. Schrauf, ‘‘EEG alpha
[31] A. Murata, ‘‘Proposal of a method to predict subjective rating on drowsi- spindle measures as indicators of driver fatigue under real traffic condi-
ness using physiological and behavioral measures,’’ IIE Trans. Occup. tions,’’ Clin. Neurophysiol., vol. 122, no. 6, pp. 1168–1178, Jun. 2011.
Ergon. Hum. Factors, vol. 4, nos. 2–3, pp. 128–140, 2016. [53] G. Li and W.-Y. Chung, ‘‘A context-aware EEG headset system for early
[32] L.-L. Chen, Y. Zhao, J. Zhang, and J.-Z. Zou, ‘‘Automatic detection detection of driver drowsiness,’’ Sensors, vol. 15, no. 8, pp. 20873–20893,
of alertness/drowsiness from physiological signals using wavelet-based Aug. 2015.
nonlinear features and machine learning,’’ Expert Syst. Appl., vol. 42, [54] L. Yang, R. Ma, H. M. Zhang, W. Guan, and S. Jiang, ‘‘Driving behavior
no. 21, pp. 7344–7355, Nov. 2015. recognition using EEG data from a simulated car-following experiment,’’
[33] A. Sahayadhas, K. Sundaraj, M. Murugappan, and R. Palaniappan, Accident Anal. Prevention, vol. 116, pp. 30–40, Jul. 2018.
‘‘Physiological signal based detection of driver hypovigilance using [55] S. P. A. Drummond, J. C. Gillin, and G. G. Brown, ‘‘Increased cerebral
higher order spectra,’’ Expert Syst. Appl., vol. 42, no. 22, pp. 8669–8677, response during a divided attention task following sleep deprivation,’’
Dec. 2015. J. Sleep Res., vol. 10, no. 2, pp. 85–92, 2001.
[34] A. Sahayadhas, K. Sundaraj, M. Murugappan, and R. Palaniappan, [56] M. D. Manzar, M. Salahuddin, M. Alamri, A. Albougami, M. Y. A. Khan,
‘‘A physiological measures-based method for detecting inattention in D. Nureye, D. W. Spence, and S. R. Pandi-Perumal, ‘‘Psychometric prop-
drivers using machine learning approach,’’ Biocybern. Biomed. Eng., erties of the Epworth sleepiness scale in Ethiopian university students,’’
vol. 35, no. 3, pp. 198–205, 2015. Health Qual. Life Outcomes, vol. 17, Feb. 2019, Art. no. 30.
[35] Z. Zhang, T.-P. Jung, S. Makeig, Z. Pi, and B. D. Rao, ‘‘Spatiotemporal [57] Y. Krishnamoorthy, G. Sarveswaran, M. Sakthivel, A. Kalaiselvy,
sparse Bayesian learning with applications to compressed sensing of M. G. Majella, and S. Lakshminarayanan, ‘‘Construct validation and
multichannel physiological signals,’’ IEEE Trans. Neural Syst. Rehabil. reliability assessment of tamil version of epworth sleepiness scale to
Eng., vol. 22, no. 6, pp. 1186–1197, Nov. 2014. evaluate daytime sleepiness among adolescents in rural puducherry,
[36] M. Akhlaq, T. R. Sheltami, B. Helgeson, and E. M. Shakshuki, ‘‘Design- South India,’’ J. Neurosci. Rural Pract., vol. 10, no. 1, pp. 89–93,
ing an integrated driver assistance system using image sensors,’’ J. Intell. Jan./Mar. 2019.
Manuf., vol. 23, no. 6, pp. 2109–2132, Dec. 2012. [58] J. Stonehouse, A. Turton, D. Mansfield, and G. S. Hamilton, ‘‘A novel
quality assurance program for multiple sleep latency tests,’’ J. Sleep Res.,
[37] S.-J. Jung, H.-S. Shin, and W.-Y. Chung, ‘‘Driver fatigue and drowsiness
vol. 27, p. e85_12766, Oct. 2018.
monitoring system with embedded electrocardiogram sensor on steering
wheel,’’ IET Intell. Transp. Syst., vol. 8, no. 1, pp. 43–50, Feb. 2014. [59] E. J. Denton, M. Barnes, T. Churchward, M. Jackson, A. Collins,
M. T. Naughton, and E. Dabscheck, ‘‘Mood disorders are highly prevalent
[38] A. Reyes-Muñoz, M. C. Domingo, M. A. López-Trinidad, and
in patients investigated with a multiple sleep latency test,’’ Sleep Breath-
J. L. Delgado, ‘‘Integration of body sensor networks and vehicular ad-hoc
ing, vol. 22, no. 2, pp. 305–309, May 2018.
networks for traffic safety,’’ Sensors, vol. 16, no. 1, p. 107, Jan. 2016.
[60] S. A. K. Dahi, W. A. G. A. Nasser, A. M. Almoabdi, S. A. M. Asiri,
[39] S. Vitabile, A. De Paola, and F. Sorbello, ‘‘A real-time non-intrusive
N. O. S. Bazhair, R. A. A. Abdullah, A. A. Mohmmed, B. A. Alsultan,
FPGA-based drowsiness detection system,’’ J. Ambient Intell. Humanized
F. A. Alhawaj, A. G. AlHashim, and N. S. N. Alasmari, ‘‘Management
Comput., vol. 2, no. 4, pp. 251–262, Dec. 2011.
of sleep disorder in family practice,’’ Indo Amer. J. Pharmaceutical Sci.,
[40] S. Ahn, T. Nguyen, H. Jang, J. G. Kim, and S. C. Jun, ‘‘Exploring
vol. 6, no. 1, pp. 1571–1576, 2019.
neuro-physiological correlates of drivers’ mental fatigue caused by sleep
[61] Y.-S. Huang, C. Guilleminault, C.-H. Lin, C.-H. Chen, W.-C. Chin,
deprivation using simultaneous EEG, ECG, and fNIRS data,’’ Frontiers
and T.-S. Chen, ‘‘Multiple sleep latency test in narcolepsy type 1 and
Hum. Neurosci., vol. 10, p. 219, May 2016.
narcolepsy type 2: A 5-year follow-up study,’’ J. Sleep Res., vol. 27, no. 5,
[41] J. C. W. Brooks, O. K. Faull, K. T. S. Pattinson, and M. Jenkinson, Oct. 2018, Art. no. UNSP e12700.
‘‘Physiological noise in brainstem FMRI,’’ Frontiers Hum. Neurosci., [62] D. T. Plante, J. D. Cook, and M. L. Prairie, ‘‘Multimodal hypersom-
vol. 7, p. 623, Oct. 2013. nolence assessment substantially increases objective identification of
[42] L. Xu, B. Wang, G. Xu, W. Wang, Z. Liu, and Z. Li, ‘‘Functional hypersomnolence in patients referred for multiple sleep latency testing,’’
connectivity analysis using fNIRS in healthy subjects during prolonged Sleep, vol. 41, pp. A227–A228, Apr. 2018.
simulated driving,’’ Neurosci. Lett., vol. 640, pp. 21–28, Feb. 2017. [63] T. B. Young and H. C. Kim, ‘‘Differential effects of sleepiness, measured
[43] S.-K. Yeom, D.-O. Won, S. I. Chi, K.-S. Seo, H. J. Kim, K.-R. Müller, by multiple sleep latency test (MSLT) and Stanford sleepiness scale
and S.-W. Lee, ‘‘Spatio-temporal dynamics of multimodal EEG-fNIRS (SSS), on psychomotor vigilance test (PVT),’’ Sleep, vol. 26, p. A439,
signals in the loss and recovery of consciousness under sedation using May 2003.
midazolam and propofol,’’ PLoS ONE, vol. 12, no. 11, Nov. 2017, [64] E. Hoddes, V. Zarcone, and W. Dement, ‘‘Development and use of
Art. no. e0187743. stanford sleepiness scale (SSS),’’ Psychophysiology, vol. 9, no. 1, p. 150,
[44] P. Wang, J. Min, and J. Hu, ‘‘Ensemble classifier for driver’s fatigue 1972.
detection based on a single EEG channel,’’ IET Intell. Transp. Syst., [65] J. Herscovitch and R. Broughton, ‘‘Sensitivity of the Stanford sleepiness
vol. 12, no. 10, pp. 1322–1328, Dec. 2018. scale to the effects of cumulative partial sleep deprivation and recovery
[45] P. Forsman, I. Pyykkö, E. Toppila, and E. Hæggström, ‘‘Feasibility of oversleeping,’’ Sleep, vol. 4, no. 1, pp. 83–92, 1981.
force platform based roadside drowsiness screening—A pilot study,’’ [66] F. Regen, H. Dorn, and H. Danker-Hopfe, ‘‘The relationship between
Accident Anal. Prevention, vol. 62, pp. 186–190, Jan. 2014. pupillography and the subjective assessment of sleepiness with the
[46] J. Jo, S. J. Lee, J. Kim, H. G. Jung, and K. R. Park, ‘‘Vision-based method Stanford sleepiness scale,’’ Sleep, vol. 26, p. A408, May 2003.
for detecting driver drowsiness and distraction in driver monitoring sys- [67] K. Dos Santos, F. M. Fischer, D. F. de Andrade, P. A. Barbetta,
tem,’’ Opt. Eng., vol. 50, no. 12, Dec. 2011, Art. no. 127202. M. R. A. Cardoso, and R. A. D. Vilela, ‘‘Shift work and sex are variants
[47] K. Karthick, T. Praveen, R. Yogalakshmi, R. Anushya, and M. H. Meha- across Karolinska sleep questionnaire and the application of linking
jabeen, ‘‘Accident prevention in vehicle with efective rescue operation,’’ analysis as a step forward using item response theory,’’ Chronobiol. Int.,
Iioab J., vol. 7, no. 9, pp. 1–5, Aug. 2016. vol. 36, no. 1, pp. 42–56, Jan. 2019.
[48] W. B. Verwey and D. M. Zaidel, ‘‘Preventing drowsiness accidents by an [68] T. Åkerstedt, D. Hallvig, and G. Kecklund, ‘‘Normative data on the diur-
alertness maintenance device,’’ Accident Anal. Prevention, vol. 31, no. 3, nal pattern of the Karolinska Sleepiness Scale ratings and its relation to
pp. 199–211, May 1999. age, sex, work, stress, sleep quality and sickness absence/illness in a large
[49] Drowsy Driving and Automobile Crashes, National Center Sleep Disorder sample of daytime workers,’’ J. Sleep Res., vol. 26, no. 5, pp. 559–566,
Res. Nat. Highway Traffic Safety Admin., Howe, TX, USA, 1998. Oct. 2017.

167184 VOLUME 7, 2019

M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

[69] T. Akerstedt, K. Hume, D. Minors, and J. Waterhouse, ‘‘The sub- [91] F. Senny, J. Destine, and R. Poirrier, ‘‘Midsagittal jaw movement analysis
jective meaning of good sleep, an intraindividual approach using the for the scoring of sleep Apneas and Hypopneas,’’ IEEE Trans. Biomed.
Karolinska Sleep Diary,’’ Perceptual Motor Skills, vol. 79, no. 1, Eng., vol. 55, no. 1, pp. 87–95, Jan. 2008.
pp. 287–296, Aug. 1994. [92] T. Arima, A. Tomonaga, M. Toyota, S.-I. Inoue, N. Ohata, and
[70] K. Otsuka, M. Ikari, Y. Ichimaru, H. Saito, T. Kawakami, K. Otsuka, P. Svensson, ‘‘Does restriction of mandibular movements during sleep
H. Kaba, and K. Seto, ‘‘Experimental study on the relationship between influence jaw-muscle activity?’’ J. Oral Rehabil., vol. 39, no. 7,
cardiac arrhythmias and sleep states by ambulatory ecg-EEC monitor- pp. 545–551, Jul. 2012.
ing,’’ Clin. Cardiol., vol. 9, no. 7, pp. 305–313, 1986. [93] G. Barbato, G. Ficca, M. Beatrice, M. Casiello, G. Muscettola, and
[71] L. M. Makarov, M. A. Shkolnikova, V. V. Bereznitskaya, and F. Rinaldi, ‘‘Effects of sleep deprivation on spontaneous eye blink rate
R. R. Giorgobiani, ‘‘Cardiac rhythm in sleeping healthy adolescents as and alpha EEG power,’’ Biol. Psychiatry, vol. 38, no. 5, pp. 340–341,
recorded by Holter ECG monitoring,’’ Terapevticheskii Arkhiv, vol. 71, Sep. 1995.
no. 1, pp. 19–22, 1999. [94] M. D. D. Ebert, R. D. R. Albert, G. Hammon, B. Strasser, M. D. A. May,
[72] I. Stamatoiu, H. Guiloni, and V. Gabersek, ‘‘Study of the respiratory and M. D. A. Merz, ‘‘Eye-blink rates and depression: Is the antidepressant
frequency during sleep in relation to the amplitude of the R wave of the effect of sleep deprivation mediated by the dopamine system?’’ Neuropsy-
ECG,’’ Electroencephalogr. Clin. Neurophysiol., vol. 29, no. 1, p. 100, chopharmacology, vol. 15, no. 4, pp. 332–339, Oct. 1996.
1970. [95] P. P. Caffier, U. Erdmann, and P. Ullsperger, ‘‘The spontaneous eye-
[73] O. Tímuş and E. D. Bolat, ‘‘k-NN-based classification of sleep apnea blink as sleepiness indicator in patients with obstructive sleep apnoea
types using ECG,’’ Turkish J. Electr. Eng. Comput. Sci., vol. 25, no. 4, syndrome-a pilot study,’’ Sleep Med., vol. 6, no. 2, pp. 155–162,
pp. 3008–3023, 2017. Mar. 2005.
[74] F. Ebrahimi, S. K. Setarehdan, R. Martinek, and H. Nazeran, ‘‘Compar- [96] M. A. Kamran, M. M. N. Mannan, and M. Y. Jeong, ‘‘Corti-
ative analysis of the discriminative capacity of EEG, two ECG-derived cal signal analysis and advances in functional near-infrared spec-
and respiratory signals in automatic sleep staging,’’ Adv. Electr. Electron. troscopy signal: A review,’’ Frontiers Hum. Neurosci., vol. 10, p. 261,
Eng., vol. 15, no. 3, pp. 459–475, Sep. 2017. Jun. 2016.
[75] N. Zhang, M. Fard, M. H. U. Bhuiyan, D. Verhagen, M. F. Azari, and [97] T. Nguyen, S. Ahn, H. Jang, S. C. Jun, and J. G. Kim, ‘‘Utilization of
S. R. Robinson, ‘‘The effects of physical vibration on heart rate variability a combined EEG/NIRS system to predict driver drowsiness,’’ Sci. Rep.,
as a measure of drowsiness,’’ Ergonomics, vol. 61, no. 9, pp. 1259–1272, vol. 7, Mar. 2017, Art. no. 43933.
2018. [98] C. Dissanayaka, E. Ben-Simon, M. Gruberger, A. Maron-Katz, H. Sharon,
[76] J. Vicente, P. Laguna, A. Bartra, and R. Bailon, ‘‘Drowsiness detection T. Hendler, and D. Cvetkovic, ‘‘Comparison between human awake, med-
using heart rate variability,’’ Med. Biol. Eng. Comput., vol. 54, no. 6, itation and drowsiness EEG activities based on directed transfer function
pp. 927–937, Jun. 2016. and MVDR coherence methods,’’ Med. Biol. Eng. Comput., vol. 53, no. 7,
[77] G. Li and W.-Y. Chung, ‘‘Detection of driver drowsiness using pp. 599–607, Jul. 2015.
wavelet analysis of heart rate variability and a support vector
[99] M. M. N. Mannan, M. Y. Jeong, and M. A. Kamran, ‘‘Hybrid
machine classifier,’’ Sensors, vol. 13, no. 12, pp. 16494–16511,
ICA—Regression: Automatic identification and removal of ocular arti-
Dec. 2013.
facts from electroencephalographic signals,’’ Frontiers Hum. Neurosci.,
[78] R. Hilbert and P. Naitoh, ‘‘EOG and delta rhythmicity in human sleep vol. 10, p. 193, May 2016.
EEG,’’ Psychophysiology, vol. 9, no. 5, pp. 533–538, 1972.
[100] M. M. N. Mannan, S. Kim, M. Y. Jeong, and M. A. Kamran, ‘‘Hybrid
[79] M. M. Rahman, M. I. H. Bhuiyan, and A. R. Hassan, ‘‘Sleep stage
EEG—Eye tracker: Automatic identification and removal of eye move-
classification using single-channel EOG,’’ Comput. Biol. Med., vol. 102,
ment and blink artifacts from electroencephalographic signal,’’ Sensors,
pp. 211–220, Nov. 2018.
vol. 16, no. 2, p. 241, Feb. 2016.
[80] M. Dursun, S. Özşen, C. Yücelbaş, Ş. Yücelbaş, G. Tezel, S. Küççüktürk,
[101] M. M. N. Mannan, M. A. Kamran, S. Kang, and M. Y. Jeong, ‘‘Effect
and Ş. Yosunkaya, ‘‘A new approach to eliminating EOG artifacts from
of EOG signal filtering on the removal of ocular artifacts and EEG-
the sleep EEG signals for the automatic sleep stage classification,’’ Neural
based brain-computer interface: A comprehensive study,’’ Complexity,
Comput. Appl., vol. 28, no. 10, pp. 3095–3112, Oct. 2017.
vol. 2018, Jul. 2018, Art. no. 4853741.
[81] S.-F. Liang, C.-E. Kuo, Y.-C. Lee, W.-C. Lin, Y.-C. Liu, P.-Y. Chen,
F.-Y. Cherng, and F.-Z. Shaw, ‘‘Development of an EOG-based automatic [102] M. A. Kamran, M. Y. Jeong, and M. Mannan, ‘‘Optimal hemodynamic
sleep-monitoring eye mask,’’ IEEE Trans. Instrum. Meas., vol. 64, no. 11, response model for functional near-infrared spectroscopy,’’ Frontiers
pp. 2977–2985, Nov. 2015. Behav. Neurosci., vol. 9, p. 151, Jun. 2015.
[82] G. Gruber, S. Parapatics, E. Loretz, S. Schmittner, and G. Dorffner, [103] M. M. N. Mannan, M. A. Kamran, and M. Y. Jeong, ‘‘Identification and
‘‘Validity of sleep staging based on a reduced montage using 2 EOG removal of physiological artifacts from electroencephalogram signals:
channels,’’ J. Sleep Res., vol. 25, p. 344, Sep. 2016. A review,’’ IEEE Access, vol. 6, pp. 30630–30652, 2018.
[83] Q. R. Regestein, E. Hartmann, and P. Reich, ‘‘Head movement disorder [104] R. S. Menon and S.-G. Kim, ‘‘Spatial and temporal limits in cognitive
occurring in dreaming sleep,’’ J. Nervous Mental Disease, vol. 164, no. 6, neuroimaging with fMRI,’’ Trends Cogn. Sci., vol. 3, no. 6, pp. 207–216,
pp. 432–436, 1977. 1999.
[84] P. Gagnon and J. De koninck, ‘‘Repetitive head movements during REM [105] R. C. deCharms, ‘‘Applications of real-time fMRI,’’ Nature Rev. Neu-
sleep,’’ Biol. Psychiatry, vol. 20, no. 2, pp. 176–178, 1985. rosci., vol. 9, no. 9, pp. 720–729, 2008.
[85] C. McGuigan, M. Lunn, and M. C. Walker, ‘‘Dopamine agonist respon- [106] C. Bulgarelli, A. Blasi, S. Arridge, S. Powell, C. C. J. M. de Klerk,
sive periodic head movements in sleep-an unusual adult-onset parasom- V. Southgate, S. Brigadoi, W. Penny, S. Tak, and A. Hamilton, ‘‘Dynamic
nia,’’ Movement Disorders, vol. 21, p. S429, 2006. causal modelling on infant fNIRS data: A validation study on a
[86] M. Kohsaka et al., ‘‘Head jerk during REM sleep: Healthy motor event or simultaneously recorded fNIRS-fMRI dataset,’’ NeuroImage, vol. 175,
movement disorder?’’ J. Sleep Res., vol. 27. Hoboken, NJ, USA: Wiley, pp. 413–424, Jul. 2018.
2018. [107] C. Chang and G. H. Glover, ‘‘Time-frequency dynamics of resting-state
[87] K. N. Anderson, I. E. Smith, and J. M. Shneerson, ‘‘Rhythmic brain connectivity measured with fMRI,’’ NeuroImage, vol. 50, no. 1,
movement disorder (head banging) in an adult during rapid eye pp. 81–98, 2010.
movement sleep,’’ Movement Disorders, vol. 21, no. 6, pp. 866–867, [108] E. Tagliazucchi and H. Laufs, ‘‘Decoding wakefulness levels from typical
Jun. 2006. fMRI resting-state data reveals reliable drifts between wakefulness and
[88] F. Senny, J. Destiné, and R. Poirrier, ‘‘Midsagittal jaw movements sleep,’’ Neuron, vol. 82, no. 3, pp. 695–708, 2014.
as a sleep/wake marker,’’ IEEE Trans. Biomed. Eng., vol. 56, no. 2, [109] E. Tagliazucchi, F. Von Wegner, A. Morzelewski, V. Brodbeck,
pp. 303–309, Feb. 2009. S. Borisov, K. Jahnke, and H. Laufs, ‘‘Large-scale brain functional
[89] R. Poirrier, L. Cambron, P. Maquet, J. Destine, and F. Senny, ‘‘Detection modularity is reflected in slow electroencephalographic rhythms across
of sleep and wake stages based on jaw movement analysis,’’ J. Sleep Res., the human non-rapid eye movement sleep cycle,’’ NeuroImage, vol. 70,
vol. 17, p. 233, Dec. 2008. pp. 327–339, Apr. 2013.
[90] Y. Ruckebusch, R. W. Dougherty, and H. M. Cook, ‘‘Jaw movements and [110] T. Liu, ‘‘Positive correlation between drowsiness and prefrontal activation
rumen motility as criteria for measurement of deep sleep in cattle,’’ Amer. during a simulated speed-control driving task,’’ NeuroReport, vol. 25,
J. Vet. Res., vol. 35, no. 10, pp. 1309–1312, 1974. no. 16, pp. 1316–1319, 2014, doi: 10.1097/WNR.0000000000000265.

VOLUME 7, 2019 167185

 M. A. Kamran et al.: Drowsiness, Fatigue and Poor Sleep’s Causes and Detection: Comprehensive Study

[111] N.-H. Liu, C.-Y. Chiang, and H.-M. J. S. Hsu, ‘‘Improving driver alertness [124] A. E. Shortz, A. Pickens, Q. Zheng, and R. K. Mehta, ‘‘The effect of
through music selection using a mobile EEG to detect brainwaves,’’ cognitive fatigue on prefrontal cortex correlates of neuromuscular fatigue
Sensors, vol. 13, no. 7, pp. 8199–8221, 2013. in older women,’’ J. Neuroeng. Rehabil., vol. 12, p. 115, Dec. 2015.
[112] N. Gurudath and H. B. Riley, ‘‘Drowsy driving detection by EEG analysis [125] E. A. Allen, E. Damaraju, T. Eichele, L. Wu, and V. D. Calhoun, ‘‘EEG
using wavelet transform and K-means clustering,’’ Proc. Comput. Sci., signatures of dynamic functional network connectivity states,’’ Brain
vol. 34, pp. 400–409, 2014. Topography, vol. 31, no. 1, pp. 101–116, Jan. 2018.
[113] J. Chen, H. Wang, and C. Hua, ‘‘Assessment of driver drowsiness using [126] A. Haimovici, E. Tagliazucchi, P. Balenzuela, and H. Laufs, ‘‘On wakeful-
electroencephalogram signals based on multiple functional brain net- ness fluctuations as a source of BOLD functional connectivity dynamics,’’
works,’’ Int. J. Psychophysiol., vol. 133, pp. 120–130, Nov. 2018. Sci. Rep., vol. 7, Jul. 2017, Art. no. 5908.
[114] H. Wang, A. Dragomir, N. I. Abbasi, J. Li, N. V. Thakor, and [127] J. L. Ong, D. Y. Kong, T. T. Y. Chia, J. Tandi, B. T. T. Yeo,
A. Bezerianos, ‘‘A novel real-time driving fatigue detection system based and M. W. L. Chee, ‘‘Co-activated yet disconnected—Neural correlates
on wireless dry EEG,’’ Cogn. Neurodyn. vol. 12, no. 4, pp. 365–376, of eye closures when trying to stay awake,’’ NeuroImage, vol. 118,
2018. pp. 553–562, Sep. 2015.
[115] C. Zhang, W. Wang, C. Chen, C. Zeng, D. E. Anderson, and B. Cheng, [128] G. R. Poudel, C. R. H. Innes, P. J. Bones, R. Watts, and R. D. Jones, ‘‘Los-
‘‘Determination of optimal electroencephalography recording locations ing the struggle to stay awake: Divergent thalamic and cortical activity
for detecting drowsy driving,’’ IET Intell. Transp. Syst., vol. 12, no. 5, during microsleeps,’’ Hum. Brain Mapping, vol. 35, no. 1, pp. 257–269,
pp. 345–350, Jun. 2018. Jan. 2014.
[116] T. Li, Y. Lin, Y. Gao, and F. Zhong, ‘‘Longtime driving induced cere- [129] H. van Duinen, R. Renken, N. Maurits, and I. Zijdewind, ‘‘Effects of
bral hemodynamic elevation and behavior degradation as assessed by motor fatigue on human brain activity, an fMRI study,’’ NeuroImage,
functional near-infrared spectroscopy and a voluntary attention test,’’ vol. 35, no. 4, pp. 1438–1449, May 2007.
J. Biophoton., vol. 11, no. 12, Dec. 2018, Art. no. e201800160. [130] M. Maire, C. F. Reichert, V. Gabel, A. U. Viola, C. Phillips, C. Berthomier,
[117] G. Borragán, M. Gilson, C. Guerrero-Mosquera, E. Di Ricci, H. Slama, S. Borgwardt, C. Cajochen, and C. Schmidt, ‘‘Human brain patterns
and P. Peigneux, ‘‘Transcranial direct current stimulation does not coun- underlying vigilant attention: Impact of sleep debt, circadian phase and
teract cognitive fatigue, but induces sleepiness and an inter-hemispheric attentional engagement,’’ Sci. Rep., vol. 8, no. 1, p. 970, 2018.
shift in brain oxygenation,’’ Frontiers Psychol., vol. 9, Nov. 2018, [131] F. Gharagozlou, G. N. Saraji, A. Mazloumi, A. Nahvi, A. M. Nasrabadi,
Art. no. 2351. A. R. Foroushani, A. A. Kheradmand, M. Ashouri, and M. Samavati,
[118] T. Cai, H. Zhu, J. Xu, S. Wu, X. Li, and S. He, ‘‘Human cortical neural ‘‘Detecting driver mental fatigue based on EEG alpha power changes
correlates of visual fatigue during binocular depth perception: An fNIRS during simulated driving,’’ Iranian J. Public Health, vol. 44, no. 12,
study,’’ PLoS ONE, vol. 12, no. 2, Feb. 2017, Art. no. e0172426. p. 1693, 2015.
[119] C.-H. Chuang, Z. Cao, J.-T. King, B.-S. Wu, Y.-K. Wang, and C.-T. Lin, [132] J. Chen, H. Wang, C. Hua, Q. Wang, and C. Liu, ‘‘Graph analysis of
‘‘Brain electrodynamic and hemodynamic signatures against fatigue dur- functional brain network topology using minimum spanning tree in driver
ing driving,’’ Frontiers Neurosci., vol. 12, p. 181, Mar. 2018. drowsiness,’’ Cogn. Neurodyn., vol. 12, no. 6, pp. 569–581, Dec. 2018.
[120] M. Muthalib, B. Kan, K. Nosaka, and S. Perrey, ‘‘Effects of transcranial [133] M. Ramzan, H. U. Khan, S. M. Awani, A. Ismail, M. Ilyas, and
direct current stimulation of the motor cortex on prefrontal cortex activa- A. Mahmood, ‘‘A survey on state-of-the-art drowsiness detection tech-
tion during a neuromuscular fatigue task: An fNIRS study,’’ in Oxygen niques,’’ IEEE Access, vol. 7, pp. 61904–61919, 2019.
Transport to Tissue XXXV (Advances in Experimental Medicine and [134] C. I. Karageorghis, M. Bigliassi, S. M. R. Guérin, and Y. Delevoye-
Biology), vol. 789, S. VanHuffel, G. Naulaers, A. Caicedo, D. F. Bruley, Turrell, ‘‘Brain mechanisms that underlie music interventions in the
and D. K. Harrison, Eds. New York, NY, USA: Springer, 2013, exercise domain,’’ Prog. Brain Res., vol. 240, pp. 109–125, 2018, doi:
pp. 73–79. 10.1016/bs.pbr.2018.09.004.
[121] G. Ochi, Y. Yamada, K. Hyodo, K. Suwabe, T. Fukuie, K. Byun, I. Dan, [135] R. K. Mehta and J. Rhee, ‘‘Age-specific neural strategies to maintain
and H. Soya, ‘‘Neural basis for reduced executive performance with motor performance after an acute social stress bout,’’ Exp. Brain Res.,
hypoxic exercise,’’ NeuroImage, vol. 171, pp. 75–83, May 2018. vol. 235, no. 7, pp. 2049–2057, Jul. 2017, doi: 10.1007/s00221-017-
[122] J. Rhee and R. K. Mehta, ‘‘Functional connectivity during handgrip 4949-9.
motor fatigue in older adults is obesity and sex-specific,’’ Frontiers Hum. [136] R. K. Mehta, ‘‘Stunted PFC activity during neuromuscular control under
Neurosci., vol. 12, p. 455, Nov. 2018. stress with obesity,’’ Eur. J. Appl. Physiol., vol. 116, no. 2, pp. 319–326,
[123] H. O. Richter, M. Forsman, G. H. Elcadi, R. Brautaset, J. E. Marsh, and Feb. 2016, doi: 10.1007/s00421-015-3283-9.
C. Zetterberg, ‘‘Prefrontal cortex oxygenation evoked by convergence [137] L. Bu, D. Wang, C. Huo, G. Xu, Z. Li, and J. Li, ‘‘Effects of poor
load under conflicting stimulus-to-accommodation and stimulus-to- sleep quality on brain functional connectivity revealed by wavelet-based
vergence eye-movements measured by NIRS,’’ Frontiers Hum. Neurosci., coherence analysis using NIRS methods in elderly subjects,’’ Neurosci.
vol. 12, p. 298, Jul. 2018. Lett., vol. 668, pp. 108–114, Mar. 2018.

167186 VOLUME 7, 2019