Physiological Data Recording in VR Simulator for Sleepiness Detection

During Driving

Chevallier Baptiste

1,2 a

, Istrate Dan

1 b

, Zalc Vincent

1 c

, Vera Nicolas

and Labrousse Christophe

2,3

Biomechanics and Bioengineering, UMR CNRS 7338, Universit

e de Technologie de Compi

egne, Compi

egne, France

CoreForTech, FIPSICO, Lille, France

PredexIA, Toulouse, France

Keywords:

Augmented Reality, Heart Rate Variability, PPG, EEG, Signal Preprocessing.

Abstract:

Drowsy driving is a major issue in road safety. In this paper, we propose a description of an experimental

data collection to develop a drowsiness detection model. The objective of this data collection was mainly to

gather physiological data of individuals in simulated driving situations. We designed a realistically annoying

scenario to induce fatigue while staying close to real driving conditions. The experiment was run on an

augmented reality platform called CAVE. The need for contextualization came early in the design of the

experiment. Therefore, in addition to physiological data, we added much more data sources, from driving

habits to driving behaviour in addition to self-assessment of fatigue levels and the gold standard (EEG). As a

result, this experience helped us create a data set of physiological data completed by elements of context and

driving behaviour. Thus allowing us to perform a very rich analysis of these physiological data.

1 INTRODUCTION

Drowsy driving is one of the deadliest causes of acci-

dents (Board, 1999). Yet it is really difﬁcult to esti-

mate the part of drowsiness among other causes of ac-

cidents because fatigue isn’t easy to measure after an

accident. While drug use can be measured in blood,

speeding can be evaluated from structure deformation

but the equivalent analysis is impossible with drowsi-

ness.

Therefore, industrial (Friedrichs and Yang, 2010)

and political (EuropeanCommission, 2021) actors are

actively trying to ﬁnd solutions to detect drowsiness

and alert drivers while he is still awake. From a

broader perspective, monitoring driver attention on

the driving task will be a major challenge for future

vehicles. Autonomous driving cars is a very popular

topic and major car companies are trying to make this

a reality. For the next generation of vehicles to come,

the autonomous level will be at 3, which means the

driver isn’t required to keep his hand on the wheel but

should be able to take back control at any time. There-

fore drowsiness and attention monitoring will be key,

https://orcid.org/0000-0003-3331-3786

https://orcid.org/0000-0001-5906-4947

https://orcid.org/0000-0001-5325-6649

as the car should ensure the driver will be able to take

that control back.

Our vehicles are becoming more of computers on

wheels. They are full of sensors gathering data on the

road, on the driver, or on the weather. Our approach

is to replicate the type of information available in a

modern car and use the combination of this data to de-

tect drowsiness. In this study, we tried to gather data

used in a multitude of systems. We record driving

behaviour-related data, video of the face, and physio-

logical signals, thus covering the three main solutions

developed to detect drowsy driving.

Doing that we set two important constraints for

our experiment. First, it should be immersive for the

subject. Being able to offer a driving environment

closer to reality will help to record more relevant data

as the subject will have close-to-reality reactions. The

second, constraint is to limit the invasiveness of sen-

sors. We thought that sensors used in the experiment

should be able to be used in real driving situations

by everyone. This is why we chose small Bluetooth

sensors which didn’t interfere with the subject driving

actions.

In this paper, we will describe the used hardware

and explain how we set up the experimental scenario.

We will then go through all the collected data and how

we processed them in order to create a clean and rich

408

Chevallier, B., Istrate, D., Zalc, V., Vera, N. and Labrousse, C.

Physiological Data Recording in VR Simulator for Sleepiness Detection During Driving.

DOI: 10.5220/0011698100003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 5: HEALTHINF, pages 408-415

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

database. We will also give the results of our ﬁrst

analysis.

The experiment took place in France, in order to

record data from subjects we have obtained an agree-

ment with the ethical research committee of Sorbonne

University (CER-SU) on protocol n°CER-2021-018.

2 RELATED WORK

Many studies have been conducted to develop drowsi-

ness detection systems (Sahayadhas et al., 2012). We

can divide them into three major categories.

The ﬁrst category is the analysis of driving be-

haviour using vehicle information like steering wheel

angle or lane departure detection. Datasets have been

recorded by car manufacturers in real driving condi-

tions (Friedrichs and Yang, 2010) but most commonly

in simulated conditions like (Chai et al., 2019). Using

the ﬁrst approach you aim to develop a system that

can easily be embedded in a vehicle. The latest ap-

proach aims to gather the maximum data possible to

ﬁnd out which are the most important signals. Mod-

els have been developed to anticipate behaviour that

might be dangerous on the road such as (Kim et al.,

2017) work. Other models use this information to

provide personalised help to the driver (Hori et al.,

2016).

The second category uses cameras to analyse

driver faces, especially eye closure or blinking like

in (McKinley et al., 2011) and (Abe et al., 2011).

The latest approaches in this domain use deep learn-

ing methods to process images and show good results

in both detection and anticipation of drowsy events

(de Naurois et al., 2019)

The third approach is based on physiological sig-

nal analysis. It may be based on Electroencephalo-

gram (EEG) and Electrooculogram (EOG) but these

approaches are tough to use on real driving condi-

tions due to vehicle vibration, static electricity and

limited embeddability of devices. Cardiac signals like

Electrocardiogram (ECG) and Photoplethysmoraphy

(PPG) also seem to carry enough information and are

easier to implement in real cars with more consumer-

friendly sensors. Additionally, PPG provide an easy

set-up and a non-intrusive driving experience com-

pared with EEG and EOG.

Paper Contribution: In this paper, we present a

multi-source dataset for drowsiness detection. Our

approach aims to bring simulated driving experiments

to a new level, not by using cutting-edge technologies

but through sensors and signals that can realistically

be measured in real driving conditions for direct use

in ecological conditions.

3 DATA COLLECTION PROCESS

3.1 Inclusion Phase

According to ethical criteria and constraints related to

the study, subjects were exclusively recruited within

Universit

e de Technologie de Compi

egne (UTC).

They were either students, professors, researchers or

members of the administration that were contacted

thanks to internal weekly communication. Subjects

were invited to read the description of the experiment

and answer a few questions about themselves. This

questionnaire helps us to ﬁlter subjects on driving ex-

perience, gender and age in order to ensure diversity

within our database. We also make sure no subjects

were taking any medication or drug which could pos-

sibly modify heart rate or heart rate variability.

Lastly, since the experimentation would take place

in front of a screen we also asked them about their

tendency to feel motion sickness. If selected, subjects

were then contacted to schedule an experiment time.

3.2 Participants

97 participants volunteered to participate in our study.

4 were rejected due to inclusion criteria. 33 were se-

lected from the remaining, the other were put on a

waiting list.

3.3 Simulation Platform

The experiment took place in a Cave Automatic Vir-

tual Environement (CAVE). CAVEs are augmented

reality platforms where a subject is surrounded by

glass panels (Cruz-Neira et al., 1993). The virtual en-

vironment is projected through these panels thanks to

projectors placed outside the cube. Subjects can move

freely within the cube while being surrounded by the

virtually created environment. In this speciﬁc experi-

ment, we placed a seat with a wheel and pedals, thus

reproducing the usual driving commands. The pro-

jected environment reproduced the point of view from

the driver’s seat of a car, showing the car structure,

infotainment system and the road through the wind-

screen.

3.4 Sensors

For data collection, we used sensors available on the

shelf. Three devices were selected to study their

Physiological Data Recording in VR Simulator for Sleepiness Detection During Driving

409

Figure 1: Subject being installed in the CAVE.

ability to provide a good signal in driving situations.

These devices measured three signals: Heart Rate

Variability (HRV) with ECG, HRV with PPG and

EEG.

EEG is recorded using Dreem headband (Arnal

et al., 2019). This headband was developed to analyse

sleep quality for daily usage. It is, therefore, a very

light and easy-to-wear sensor. It records a 6 channel

EEG at 250Hz. The device shape and electrode place-

ment are described in ﬁgure 2.

Figure 2: Electrode placement for Dreem EEG sensor.

We used two different techniques to measure

HRV. The ﬁrst sensor is an elastic band with elec-

trodes attached to it with the ability to communicate

via Bluetooth. The interval identiﬁcation technique

is based on ECG and consists of the identiﬁcation of

the R peak in QRS complexes, to then measure the

time between two peaks. The identiﬁcation is per-

formed directly on the device. We used Cardiosport

TP5 Heart Rate Monitor (Cardiosport, 2022) to per-

form that measure. This device has an integrated

Bluetooth processor that can send data in live con-

ditions. The second technique uses PPG to measure

the time between two blood waves by analysing the

opacity of the veins (Allen, 2007). We used Garmin

Venu SQ (Garmin, 2022) which is a smartwatch made

to be used all day long with light and comfortable ma-

terials. This device connects to a mobile application

to send live HRV data.

3.5 Simulation Scenario

The simulation scenario proposed to all subjects was

identical. It consists of a ﬁfty kilometres left-hand

loop starting in an urban zone, then going through a

forest with varying tree density. Along the road, var-

ious left and right turns were created to reduce the

feeling of going in circles. Subjects were alone on

their side of the road to avoid unwanted collisions

with other cars that may force the simulation to stop

too early. On the other side of the road, groups of

cars were placed to improve the immersion and re-

duce the emptiness of the environment. Subjects were

asked to drive when possible with cruise control acti-

vated and set at 100km/h. They had a button on the

steering wheel speciﬁcally dedicated to that function.

Two events were added to the scenario to measure the

subject’s reaction in drowsy conditions. A previous

study proved this approach to be efﬁcient to induce

and measure stress (Zontone et al., 2020). The ﬁrst

event was placed after ﬁfteen minutes of driving. We

choose this moment as we observed in precedent mea-

sures that drivers tend to have their ﬁrst drowsy event

after 15 min on task. The event is composed of a set

of trucks and lights, simulating road work on the right

side, forcing the driver to change lanes to avoid the

collision. The second event is an animal crossing the

road. With no lights and no previous indications, this

object is harder to anticipate and requires much more

attention. It’s placed after 45 minutes of experience

in the middle of a straight line, which is the most an-

noying section of the circuit.

For the simulated environment, we choose to sim-

ulate clear weather at night just before dusk so that

rain doesn’t impact the experiment.

3.6 Experimental Protocol

Here is how a standard measure goes:

1. Welcome the subject on the platform.

2. Ask the subject to ﬁll in a questionnaire on his

sleeping habits.

HEALTHINF 2023 - 16th International Conference on Health Informatics

410

3. Start and install physiological sensors on the sub-

ject.

4. Install the subject in the CAVE.

5. Start a training loop until the subject feels at ease

with the commands.

6. Last brieﬁng on what is expected from the subject

during the experiment.

7. Auto-evaluation formation, explain what is ex-

pected from them in order to record KSS.

8. Synchronisation of all signals.

9. Driving can start.

10. After completing the two loops, the driving stops.

11. Subject is invited to leave the CAVE

12. Subject is asked to ﬁll in the last questionnaire on

his feeling during the experiment.

13. Sensors are removed.

14. End of the measure.

A simpliﬁed and schematic version of the protocol

is represented in Figure 3.

For the synchronisation part, we asked subjects to

hold their hand in front of the camera, blocking the

view, while turning the steering wheel on the right.

Meanwhile, the operator was on the computer doing

a countdown from 3 to 0. At 0, the subject releases

the steering wheel and unblocks the camera view,

and the operator presses the starting key at the same

time. Software developed for the experiment record

the timestamp of the key press.

In order to ensure synchronisation between sys-

tems that could have a small variation in clock we

used UNIX timestamps from a remote server on ev-

ery device.

As a result, we have a hand movement on the cam-

era, a common timestamp between the computer and

smartphone combined with a peak in simulator signal

due to steering at the same time. This is our synchro-

nisation point.

3.7 Personal Context

During the inclusion process and before the experi-

ment, subjects were asked to answer a few questions

about themselves. There were two main subjects cov-

ered by these questions, the ﬁrst one being drivings

habits and the second one being sleep health.

We selected only people that obtained their driv-

ing license at least one year ago and who are driv-

ing frequently since. We asked subjects more precise

questions about their driving habits regarding long

and night trips when drowsiness is more likely to ap-

pear. Regarding the sleep health topic, we asked sub-

jects to describe their sleep quality, and how many

hours of sleep they had the days before.

These data will help to detect trends amongst the

measured population. The ﬁrst analysis shows that

experienced drivers, which often drive for a long time

are less affected by the simulation. It’s also really use-

ful to analyse the correlation between the sleepiness

of a subject and the length of his previous nights.

3.8 Physiological Data

Physiological sensor record two different types of in-

formation. Firstly we have EEG signals measured on

eight channels with a sample rate of 250Hz. EEG is

the gold standard signal in sleep scoring. This signal

will be used for data annotation and analysis.

Secondly, we have two sensors gathering heart-

related data. More speciﬁcally, these sensors measure

beat-to-beat intervals.

The frequency of these data is variable as both

these sensors send new data every time they detect

a new beat. This data will later be used to develop a

classiﬁcation model.

3.9 Driving Behaviour

The simulation software is able to record many sig-

nals while the simulation progresses. We choose to

ﬁlter signals that were relevant to our simulation sce-

nario, as well as signals that proved their importance

in the literature.

The set of signals is composed of:

• Position of the Centre Of Gravity (COG) of the

vehicle on the map.

• Speed of the COG of the vehicle.

• Acceleration of the COG of the vehicle.

• Wheel angle and rotation speed.

• Pressure on the brake and acceleration pedals.

• Activation state of cruise control.

• Distance between COG centre of the nearest lane.

These signals were recorded with a sampling fre-

quency of 100 Hz.

3.10 Auto-Evaluation

Fatigue is a continuous state which isn’t perceived the

same by everyone. It is a physiological reality but

also a personal feeling. Depending on whether we

feel physical fatigue or psychological fatigue, our per-

ception of our fatigue level is quite different and can

Physiological Data Recording in VR Simulator for Sleepiness Detection During Driving

411

Figure 3: Schematic experimental protocol.

greatly vary. To normalise the deﬁnition of drowsi-

ness levels, a scale going from 1 (Very Alert) to 9

(Very sleepy) has been determined and is used in a

vast majority of studies (Kaida et al., 2006) and is also

the reference scale for European Union (EU).

Therefore we implemented the assessment of this

scale in the experiment. As for the technical solutions

we chose to ask the participants to press a numeri-

cal key on a Numpad every ﬁve minutes. This gap

was selected according to EU technical requirements

on Karolinska Sleepiness Scale (KSS) rating protocol

(EuropeanCommission, 2021).

However, note that measuring drowsiness with

this scale has many ﬂaws. First of all, it requires the

subject to make a cognitive effort in order to answer

the question. While the scale is not overly complex it

still requires some time to adapt and understand how

to correctly rate ourselves. That is why a great part of

the brieﬁng is focused on explaining KSS. Secondly,

this auto-evaluation is, in essence, subjective. People

tend to over and/or under-estimate their drowsiness

level all the time. Therefore auto-evaluation has a bad

impact on the experiment and we have to take the re-

sult with care, taking into account potential bias.

3.11 Labelling Data

Labelling our dataset is a key part of our study since

we aim to use it for supervised machine-learning ap-

plications. The objective is to develop a system which

is able to detect drowsiness states in opposition to

awake states. It is a binary classiﬁcation problem.

We can use three different pieces of information

to label data:

1. EEG:

EEG is a gold standard signal used, in combina-

tion with EOG and ECG, for sleep scoring method

by sleep specialists. This method consists on the

analysis of 30 seconds segments of signal and the

association of a tag : sleep or no-sleep. This

tagging method requires a qualiﬁed set of profes-

sional reviewers.

While being the most precise on sleep detec-

tion, this method isn’t as efﬁcient to detect

rapid drowsy event. That is a major issue, be-

cause while driving subject won’t fall completely

asleep. Their drowsy events will more likely

be short moments of absence which aren’t easily

recognisable on polysomnographic records.

2. KSS:

KSS gives a subjective assessment of a fatigue

level. While this scale has 9 levels we can divide

them in two categories: 1 to 7 for awake and 7

to 9 for drowsy. However using this information

directly as a training target can have many ﬂaws.

As it is a subjective information we can’t be sure

if the level associated with the situation really cor-

responds to the true deﬁnition of the level. Also,

we can’t normalise levels among participants as

an eight for someone can mean a six for another.

3. Video:

Lastly, we can use the video of the face and the

point of view of the driver to identify drowsy be-

haviour and driving mistakes caused by drowsi-

ness. The method ﬁrst consists of the deﬁnition

and the selection of a list of observable events

that may be interesting to annotate. The list has

been deﬁned by studying literature and observing

subjects in driving simulators. The result is a list

of seven events which are: Yawning, Change of

position, Long blink, Face/Neck scratch, Loss of

Control, Accident and Sleep. The next step is re-

viewing conjointly both of the videos and select-

ing tags as they occur during the measure.

The reviewing process is done by two different

people. Then a third one validates every event.

This process ensures no events can be missed and

no event can be falsely tagged.

3.12 Dataset

The ﬁnal dataset content is summarised in table 1. It is

composed of three synchronised physiological signals

as well as nineteen behaviour-related signals, an auto-

assessment drowsy scale based on KSS, a video feed

HEALTHINF 2023 - 16th International Conference on Health Informatics

412

of the face of the driver and a video feed of the point

of view of the driver, lastly it is composed of personal

and driving habits data on the subject.

Each signal can be used for direct interpretation,

annotation or classiﬁcation purposes. Some examples

are presented in this table but the possible applications

are not limited to this.

4 PRELIMINARY ANALYSIS

In this section, we will perform a preliminary analysis

of our data. As our objective was to record drowsi-

ness we should ensure that we have sufﬁcient drowsy

drivers and drowsy events. Since we determined dif-

ferent methods of annotation the proportion of drowsy

drivers may slightly vary from one to another. The

proportions are pictured in ﬁgure 4.

58%

31%

11%

26%

48%

26%

Heavy fatigue

Light fatigue

No fatigue

Figure 4: Proportion of trips with drowsy drivers according

to KSS (left) and video (right).

In these graphics, Heavy fatigue corresponds to ei-

ther KSS > 6 or a video tag of clear loss of control.

Light fatigue is for 7 > KSS > 4 or a video tag of

early external signs of fatigue. No fatigue is for 5 >

KSS and no video tags. From this ﬁgure, we can con-

clude that these methods give signiﬁcantly different

results. Subjects tended to overestimate their fatigue

level and were globally able to drive even though

they graded their drowsy state as > 6. We can also

conclude we have been able to record a signiﬁcant

amount of fatigue events considering subjects were

not sleep deprived.

4.1 Driving behaviour

Early analysis of behaviour signals extracted from the

simulator show which one will be the most important.

We observed three principal features that were par-

ticularly affected during drowsy events, namely: Dis-

tance to the middle of the road, angle of the wheel,

and rotation speed of the wheel. Figure 5 shows an

example of these signals.

In this graphic, we can see three seconds where

there was no wheel activity while the car was going

away from the centre of the lane. Then the driver

starts a trajectory correction to replace the vehicle

closer to the centre.

Figure 5: Behaviour signals during drowsy events.

If we look at the maximum values reached during

drowsy and standard measures, we can see a drowsy

driver going far away from the centre line and having

spike activation of the wheel more critical than the

average driver (Table 2).

With these features from behavioural signals, we

can characterise the difference in behaviour between

two populations. The subsequent development would

improve these parameters and test their relevance in a

machine-learning model.

4.2 Physiological Signal Processing

The ﬁrst physiological data processing shows a great

quality difference between our sensors. For example,

with the PPG watch, we have an average recording

time of 3646 seconds (standard deviation of 100 s),

while the ECG belt has 2927 seconds average with a

standard deviation of 1396 s.

We discovered the Bluetooth connection between

the belt and the computer failed many times, thus cre-

ating very short records for some subjects. Thankfully

this behaviour only impacted six people.

5 POTENTIAL LIMITATIONS

During this experiment and while we performed our

ﬁrst analysis, we noted some limitations of our work

that should be disclaimed. However, the limitations

aren’t limited to the ones listed after that.

Even if ﬁnding differentiating characteristics to

identify events is essential for the analysis, we can’t

be sure these characteristics can be generalised. Our

model should apply to many people and shouldn’t be

helpful only for one type of driver.

Also, it is difﬁcult to evaluate how different the

subject’s physiological state would be in actual driv-

Physiological Data Recording in VR Simulator for Sleepiness Detection During Driving

413

Table 1: Dataset content summary.

Table 2: Difference between a drowsy and awake driver on

some signals.

Signal Drowsy Awake

Distance from lane centre (m)

(MAX then AVG on population) 6.2 2.5

Wheel rotation speed (rad.s

−1

)

(MAX then AVG on population) 8.8 6.7

ing conditions. Therefore, the principal solution

would be to record even more subjects to get the

most heterogeneous data set possible, thus covering

the most physiological responses imaginable.

Finally, the quantity of data is critical when devel-

oping a machine-learning algorithm. While having a

lot of drowsy events, this data set remains quite un-

balanced between awake and fatigued states. Differ-

ent techniques can be used to answer that issue, like

data augmentation.

6 CONCLUSIONS

In this research, we developed an experimental proto-

col to record the data from subjects in simulated driv-

ing conditions. Considering many previous studies,

we added a large variety of sources. The objective

was to induce fatigue by gathering the subject’s states

when they were about to fall asleep.

The experimental protocol has been validated with

self-assessment and video annotation, which makes

it reliable for machine learning developments. With

one-third of people having at least one tired event dur-

ing their measure, this data set presents enough pos-

itive events to be detected. In addition, the popula-

tion measured is quite diverse in terms of age, gender,

driving and sleep habits, making our data set a pow-

erful resource for developing a model.

The data set could be used differently to study the

correlation between physiological signals and drowsi-

ness in a mono-modal or multi-modal approach. It

could also be used to compare data sources between

them, for example, comparing the signal from ECG

band and Garmin smartwatch.

From the early analysis, we can ﬁnd differentiat-

ing characteristics for some drowsy events. The fol-

lowing work will determine if a predictive drowsiness

model can be developed on this database. This will be

possible by ﬁnding ways to generalise the character-

istics of our population. Developing features and im-

plementing some based on the literature is also an ex-

tensively considered approach. Another perspective

would be to use this protocol, which has proven to be

efﬁcient, on longer measures to gather more drowsy

events.

ACKNOWLEDGEMENTS

The authors would like to thank all the subjects for

their patience and time. Also, thanks to Heudyasic

laboratory, which gave us the opportunity to use their

platform and to develop the scenario as we wanted.

More speciﬁcally authors would like to personally

thank Mr Bouvet Yohan for his crucial help and com-

mitment to the realisation of the study.

HEALTHINF 2023 - 16th International Conference on Health Informatics

414

ETHICS DECLARATIONS

All volunteers gave their informed written consent

following approval and in accordance with the CER-

SU Review Board.

REFERENCES

Abe, T., Nonomura, T., Komada, Y., Asaoka, S., Sasai, T.,

Ueno, A., and Inoue, Y. (2011). Detecting deteriorated

vigilance using percentage of eyelid closure time dur-

ing behavioral maintenance of wakefulness tests. In-

ternational Journal of Psychophysiology, 82(3):269–

274.

Allen, J. (2007). Photoplethysmography and its application

in clinical physiological measurement. Physiological

measurement, 28(3):R1.

Arnal, P. J., Thorey, V., Ballard, M. E., Hernandez, A. B.,

Guillot, A., Jourde, H., Harris, M., Guillard, M.,

Van Beers, P., Chennaoui, M., et al. (2019). The

dreem headband as an alternative to polysomnography

for eeg signal acquisition and sleep staging. BioRxiv,

page 662734.

Board, S. (1999). Evaluation of us department of transporta-

tion efforts in the 1990s to address operator fatigue.

Safety Report NTSB/SR-99/01, Washington, DC.

Cardiosport (2022). Cardiosport TP5 device descrip-

tion. https://www.cardiosport.com/cardiosport-heart-

rate-monitors. Accessed: 2022-09-23.

Chai, M. et al. (2019). Drowsiness monitoring based on

steering wheel status. Transportation research part

D: transport and environment, 66:95–103.

Cruz-Neira, C., Sandin, D. J., and DeFanti, T. A. (1993).

Surround-screen projection-based virtual reality: the

design and implementation of the cave. In Pro-

ceedings of the 20th annual conference on Computer

graphics and interactive techniques, pages 135–142.

de Naurois, C. J., Bourdin, C., Stratulat, A., Diaz, E., and

Vercher, J.-L. (2019). Detection and prediction of

driver drowsiness using artiﬁcial neural network mod-

els. Accident Analysis & Prevention, 126:95–104.

EuropeanCommission (2021). European Com-

mission driver drowsiness and attention

warning. https://eur-lex.europa.eu/legal-

content/EN/ALL/?uri=CELEX:32021R1341. Ac-

cessed: 2022-09-23.

Friedrichs, F. and Yang, B. (2010). Drowsiness monitor-

ing by steering and lane data based features under

real driving conditions. In 2010 18th European Sig-

nal Processing Conference, pages 209–213. IEEE.

Garmin (2022). Garmin Venu SQ device description.

https://www.garmin.com/en-US/p/707174. Accessed:

2022-09-23.

Hori, C., Watanabe, S., Hori, T., Harsham, B. A., Hershey,

J., Koji, Y., Fujii, Y., and Furumoto, Y. (2016). Driver

confusion status detection using recurrent neural net-

works. In 2016 IEEE International Conference on

Multimedia and Expo (ICME), pages 1–6. IEEE.

Kaida, K., Takahashi, M.,

Akerstedt, T., Nakata, A., Otsuka,

Y., Haratani, T., and Fukasawa, K. (2006). Valida-

tion of the karolinska sleepiness scale against perfor-

mance and eeg variables. Clinical neurophysiology,

117(7):1574–1581.

Kim, I.-H., Bong, J.-H., Park, J., and Park, S. (2017). Pre-

diction of driver’s intention of lane change by aug-

menting sensor information using machine learning

techniques. Sensors, 17(6):1350.

McKinley, R. A., McIntire, L. K., Schmidt, R., Repperger,

D. W., and Caldwell, J. A. (2011). Evaluation of

eye metrics as a detector of fatigue. Human factors,

53(4):403–414.

Sahayadhas, A., Sundaraj, K., and Murugappan, M. (2012).

Detecting driver drowsiness based on sensors: a re-

view. Sensors, 12(12):16937–16953.

Zontone, P., Antonio, A., Bernardini, R., Linz, L. D., Pi-

ras, A., and Rinaldo, R. (2020). Supervised Learning

Techniques for Stress Detection in Car Drivers. Ad-

vances in Science, Technology and Engineering Sys-

tems Journal, 5(6):22–29.

Physiological Data Recording in VR Simulator for Sleepiness Detection During Driving

415