Experimental Design and Collection of Brain and Respiratory Data for

Detection of Driver’s Attention

Roman Mou

cek

1,2

, Luk

s Hnojsk

, Luk

s Va

reka

1,2

, Tom

s Prokop

1, 2

and Petr Br

uha

1,2

Department of Computer Science and Engineering, Faculty of Applied Sciences, University of West Bohemia,

Univerzitn

ı 8, Pilsen, Czech Republic

NTIS – New Technologies for the Information Society, Faculty of Applied Sciences, University of West Bohemia,

Univerzitn

ı 8, Pilsen, Czech Republic

Keywords:

Neuroinformatics, Brain Activity, Electroencephalography, Event Related Potentials, Respiration Rate,

Driver’s Attention, Simulated Drive, Data Validation, Deep Learning, Stacked Autoencoder.

Abstract:

Attention of drivers is very important for road safety and it is worth observing even in laboratory conditions

during a simulated drive. This paper deals with design of an experiment investigating driver’s attention, vali-

dation of collected data, and ﬁrst preprocessing and processing steps used within data analysis. Brain activity

is considered as a primary biosignal and is measured and analyzed using the techniques and methods of elec-

troencephalography and event related potentials. Respiration is considered as a secondary biosignal that is

captured together with brain activity. Validation of collected data using a stacked autoencoder is emphasized

as an important step preceding data analysis.

1 INTRODUCTION

Attention of drivers is a very important factor of road

safety. Inattentive drivers are dangerous to their sur-

roundings and cause a considerable number of acci-

dents. Since decline of attention, especially during

long rides, is natural, it is worthwhile to investigate

it even in laboratory conditions during a simulated

drive. Results from laboratory experiments can be

then used in real environment, e.g. for development of

devices maintaining driver’s attention or development

of autonomous driving systems used at ﬁrst when the

driver is tired or inattentive.

In this paper we follow experiments and stud-

ies previously provided by our neuroinformatics re-

search group and published in (Mou

cek and

richa,

2012), (Mou

cek and

Rond

ık, 2012), and (Mou

cek and

r, 2014). A pilot experiment that is in more

detail presented further in this paper partly shares

the same assumption as the already published experi-

ments. However, it differs in design and extends it by

collecting an additional biosignal - data from respi-

ration, and also by validation of collected data using

a stacked autoencoder. Within the presented experi-

ment the experimental design is proposed and the data

and metadata suitable for investigation of inﬂuence of

monotonous drive on driver’s attention during simu-

lated drive (a car simulator is used) are collected and

evaluated. Then the basic preprocessing and process-

ing steps used within data analysis are described.

When creating the experimental design the meth-

ods and techniques of electroencephalography (EEG)

and event related potentials (ERP) are used to moni-

tor and analyze brain activity of participating drivers.

Event related potentials measures use stimulation

techniques to investigate brain responses, so atten-

tion of a driver is not only affected by driving on

a monotonous track, but also tested and inﬂuenced us-

ing auditory stimulation in our case. It is considered

that the peak latency (peak latency represents a level

of driver’s attention) of the P3 component (the brain

cognitive response described in Section 2) increases

in time as the driver is more tired from monotonous

drive. However, this component has to be ﬁrst de-

tected in the collected data. Therefore, the data vali-

dation step has to be done. Besides the brain activity

the respiratory rate is also captured and its changes

(most probably its decrease) are anticipated.

University students in the role of tested subjects

participated in the experiment; the captured data were

analyzed and partially interpreted observing particu-

lar trends in them. Due to an assigned page limit,

a more detailed analysis including statistical evalua-

tion and more detailed discussion of the results is not

MouÄ ek R., HnojskÃ¡ L., VaÅ

Zeka L., Prokop T. and BrÅ

rha P.

Experimental Design and Collection of Brain and Respiratory Data for Detection of Driver’s Attention.

DOI: 10.5220/0006249504410450

In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 441-450

ISBN: 978-989-758-213-4

441

provided. Moreover, since we publicly provide the

collected data, the analysis can be provided by in-

dependent experts. We also realize that the number

of subjects participated in the experiment (15 partic-

ipants) is low for more precise statistical evaluation,

but this number seems to be sufﬁcient for a pilot ex-

periment and for the decision about meaningfulness

of the used experimental design and usability of the

collected data.

The paper is organized as follows. Section 2 gives

a short overview of basic principles of the ERP tech-

nique and assumptions related to amplitude and la-

tency of P3 component. Then some recent exper-

iments dealing with attention of drivers are brieﬂy

presented with respect to the experiments already

presented in papers (Mou

cek and

richa, 2012),

(Mou

cek and

Rond

ık, 2012), and (Mou

cek and Ko

2014). The objectives of the proposed experiment are

given in Section 3. Section 4 deals with experimen-

tal design, used hardware devices and software tools,

participants, environment, and course of the experi-

ment. The collected data and metadata are described

in Subsection 4.5. The validation of the collected data

is provided in Section 5. Sections 6 and 7 describe

data preprocessing and processing. Experimental re-

sults together with ﬁnal discussion are provided in

Section 8. The last section contains concluding re-

marks.

2 STATE OF THE ART

This section provides a necessary short description of

the ERP technique, the P3 component and the rela-

tion of P3 amplitude and P3 latency to attention. Then

an assumption dealing with respiratory data is given.

Finally, a short overview of EEG/ERP experiments

dealing with driver’s attention is presented.

Electroencephalography (EEG) and event related

potentials (ERP) techniques are used for monitor-

ing brain activity by measuring voltage changes on

a scalp surface. ERPs have two advantages com-

pared to classic behavioral methods: they are useful

for determining which stage or stages of processing

are inﬂuenced by a given experimental manipulation

(a detailed set of examples is in (Luck et al., 2000))

and they provide an online measure of the process-

ing of stimuli even when there is no behavioral re-

sponse (Luck, 2005).

The P3 component (also referred to as the P300

component) as one of the event related components

(waveforms) depends entirely on the task performed

by the subject and is not directly inﬂuenced by the

physical properties of the stimulus. It is sensitive to

a variety of global factors, such as time since the last

meal, weather, body temperature, and even day time

or the time of year (Luck, 2005). It is not known ex-

actly what the P3 component really means, it is proba-

bly related to a process called context updating (Luck,

2005).

However, there are known factors which inﬂuence

the amplitude and the latency of the P3 component.

The P3 component is sensitive to the probability of

a target stimulus. P3 amplitude increases when the

probability of the target stimulus class decreases or

when it is preceded by a greater number of non-target

stimuli. P3 amplitude is also larger when the subject

pays more attention to the task. However, it is smaller

if the subject does not know whether a given stimulus

is / is not a target.

P3 latency is associated with stimulus categoriza-

tion; if stimulus categorization is postponed, P3 la-

tency increases. However, P3 latency does not de-

pend on consequent processes (e.g. response selec-

tion). Thus P3 latency can be used to determine if

a performed experiment inﬂuences the processes of

stimulus categorization or processes related to a re-

sponse (Luck, 2005). More detailed information is

available in the literature provided above.

In our case we suppose that stimulus categoriza-

tion is inﬂuenced by driver fatigue that is related to

inattention and the time required for low-level sen-

sory processing of incoming stimuli increases with

the level of fatigue. We also consider that respiratory

rate decreases with the level of fatigue.

When omitting behavioral studies, not many ex-

periments dealing with driver’s attention during sim-

ulated drive were performed using the techniques of

electroencephalography and especially event related

potentials.

Suitability of EEG-based techniques is described

in (Schier, 2000); drivers’ activity during a driving

simulation task was recorded. As the result, an in-

crease in alpha activity was interpreted as less atten-

tional activity and a decrease as more attentional ac-

tivity. EEG data as an effective indicator to evalu-

ate driver fatigue are presented in (Li et al., 2012).

The impact of a surrogate Forward Collision Warning

System and its reliability according to the driver’s at-

tentional state by recording both behavioral and elec-

trophysiological data was presented in (Bueno et al.,

2012).

A systematic framework for measuring and un-

derstanding cognitive distraction in the automobile

was presented in (Strayer et al., 2015). Primary,

secondary, subjective, and physiological measures

were collected and integrated into a cognitive dis-

traction scale. Simultaneous recording of EEG and

HEALTHINF 2017 - 10th International Conference on Health Informatics

442

eye-tracking for investigating situation awareness and

working memory load in distracted driving was intro-

duced in (Ichiki et al., 2015).

The ERP technique was used in (Wester et al.,

2008) where the impact of secondary task perfor-

mance (an auditory oddball task) on a primary driv-

ing task (lane keeping) was investigated. The study

showed that when performing a simple secondary task

during driving, performance of the driving task and

this secondary task are both unaffected (Wester et al.,

2008).

Amplitude of the P3 component reﬂecting individ-

ual differences of navigation performance in a driving

task was investigated in (Bo et al., 2012). Two groups

of navigators with good and poor navigation perfor-

mance participated in a driving task; P3 amplitude

was measured while two types of triggers were pre-

sented (intersections and street signs). Poor naviga-

tors showed larger P3 amplitude than good navigators

on the left hemisphere, right hemisphere, the tempo-

ral, parietal and occipital sites when intersection trig-

gers were presented, and on the occipital site when

street sign triggers were presented, reﬂecting differ-

ent levels of mental resource needed to process the

spatial information between these two groups.

3 OBJECTIVES OF

EXPERIMENT

The presented experiment was designed and con-

ducted to investigate attention of drivers during sim-

ulated drive. The assumptions described in Section 2

were considered during designing and performing the

experiment and the following objectives were set:

• to construct and implement a monotonous track

where a substantial decrease of attention is sup-

posed,

• to use a car simulator located in our neuroinfor-

matics laboratory,

• to capture the brain activity of drivers during a

simulated drive,

• to use auditory stimuli during experiment to eval-

uate brain activity,

• to capture respiration rate together with drivers’

brain activity monitoring,

• to perform a pilot experiment on a group of at least

ten participants,

• to validate the collected data,

• to annotate, store and made the collected data pub-

lic,

• to compare the latency of averaged P3 compo-

nents and evaluate the respiration waveforms to

give a preliminary view if the results follow the

considerations given in Section 1 and should be

further elaborated.

4 DESIGN OF EXPERIMENT

The design of the experiment is a variant on the clas-

sic odd-ball paradigm in which presentations of se-

quences of frequent (non-target) audio/visual stimuli

are interrupted by infrequent (target) stimuli. Infre-

quent stimuli usually elicit a much larger P3 compo-

nent than frequent stimuli.

Two auditory stimuli were used to elicit the brain

activity of the participants:

• non-target stimulus S1 - the sound of car wipers,

duration time 900 ms, probability of occurrence p

= 0.8,

• target stimulus S2 - the sound of thunder, duration

time 900 ms, probability of occurrence p = 0.2,

Both these stimuli were played from the head-

phones worn by the participants during the whole sim-

ulated drive. The participant had to press the button to

react to each target stimulus. The response button was

located under the steering wheel. Although this re-

sponse event required some movement from the par-

ticipant, it did not cause undesired artifacts. The stim-

ulus onset asynchrony (SOA) was set to 3900 ms (it

means that the interstimulus interval was 3000 ms).

The following rules were applied within the experi-

mental scenario:

• at least two ﬁrst stimuli are non-targets,

• each target stimulus appears randomly with re-

spect to its probability of occurrence,

• two target stimuli cannot be sequential.

The background sound of drizzling was played

from speakers to imitate the real environment. The

speakers were located inside the car simulator behind

the driver’s seat.

The overall length of the experiment was 60 min-

utes. This length was experimentally veriﬁed as a

maximum time frame during which the participant did

not have difﬁculties (bad feelings) with an EEG cap

placed on his/her head. It was also expected that fa-

tigue would increase approximately after 30 minutes

of driving.

The experiment was divided into three driving ses-

sions, each session lasted 15 minutes and was fol-

lowed by ﬁve minutes break. The breaks served both

for relaxation of the participant (from driving and

Experimental Design and Collection of Brain and Respiratory Data for Detection of Driver’s Attention

443

watching the simulation scene) and for preventing

the participant from familiarity with presented stim-

uli (habituation to stimuli was thus limited).

4.1 Hardware Equipment

The experiment was performed in the neuoroinfor-

matics laboratory of the University of West Bohemia,

Czech Republic that was equipped with all necessary

hardware infrastructure for EEG/ERPs and respira-

tory rate recordings. The experimental car simulator

(a front part of a real Skoda Octavia car) was equipped

with the Logitech G27 wheel, accelerator, and brake.

These were connected to the computer via the USB

port.

Three computers were used: the ﬁrst one for pre-

sentation of auditory stimuli, the second one for stor-

ing recorded data, and the third one for presentation

of the track. The track was projected on the wall in

front of the car simulator. V-Amp produced by the

Brain Products company was used as an EEG ampli-

ﬁer. It served also as an input of the sensor capturing

respiratory rate.

4.2 Software Tools

The stimulation protocol was implemented in the Pre-

sentation software tool produced by Neurobehavioral

Systems, Inc (Neuro Behavioral Systems, 2014). The

sequence of stimuli was generated randomly, but it

always contained the same number of target and non-

target stimuli. All the sounds (including background

drizzling) were recorded using the Audacity software

tool and stored in the .wav format. The track was

prepared using the World Racing 2 game produced

by the Synetic Company (SYNETIC GmbH, 2014).

The same track as in (Mou

cek and

richa, 2012)

was used. The BrainVision Recorder (Brain Products,

2014) was used for recording and storing EEG/ERP

data and respiration rate. MATLAB, EEGLAB, and

ERPLAB software tools were used for processing and

analysis of experimental data.

4.3 Recording System

Common EEG caps (the 10-20 system deﬁning loca-

tions of scalp electrodes) were used depending on the

size of the participants’ heads. The reference elec-

trode was placed approximately 0,5 cm - 1 cm above

the nose and the ground electrode was placed on the

ear. The respiratory rate sensor (produced by the

Brain Products company) as well as the EEG cap were

connected to the V-Amp ampliﬁer.

4.4 Participants and Course of

Experiment

A group of 15 volunteers, university students (thir-

teen men, two women), aged 21-28, participated in

the experiment. The participants got necessary infor-

mation about the experiment in a written form in ad-

vance. Then informed consent was obtained from all

of them.

Before starting the experiment, each participant

was familiarized with basic behavioral rules during an

EEG/ERP experiment (e.g. not to use cosmetic prod-

ucts before the experimental session, or reduce eye

blinking and unnecessary movements to decrease the

number of artifacts). Then the participant was famil-

iarized with all sounds played during the experiment,

with car simulator controls, and with the track. Subse-

quently the participants were allowed to drive around

to get accustomed to the car simulator and simulated

drive.

During the experiment the experimenter was con-

trolling data recording and checking the correct be-

havior of the stimulation program. When the exper-

imental session ﬁnished, the participant left the car

simulator. Then the experimenter asked him/her to

ﬁll in the questionnaire containing questions related

to his/her feeling of fatigue during/after the drive.

4.5 Data and Metadata

EEG/ERP data were recorded with the sampling fre-

quency of 1 kHz; no ﬁlters were used during data

recording. The resulting signal was stored into three

ﬁles:

• .eeg ﬁle containing raw data,

• .vhdr ﬁle containing metadata that describe raw

data in the related .eeg ﬁle,

• .avg ﬁle containing the averaged signal around the

used stimuli.

All recorded data and collected metadata were

stored into the EEG/ERP portal (experiments No.

205-209, 225-236) (EEG/ERP Portal, 2016). These

data are publicly available for registered users (regis-

tration is free).

5 DATA VALIDATION

The data validation was based on the main objec-

tive of P3-based experiments: target and non-target

trials are expected to be associated with differently

shaped ERP components, especially P2, N2, and

HEALTHINF 2017 - 10th International Conference on Health Informatics

444

P3 (Blankertz et al., 2011). To validate this objective,

dichotomous machine learning was used. If classi-

ﬁcation of a speciﬁc dataset from one subject yields

low error rates (deﬁned later), the objective of the

odd-ball paradigm is considered to be fulﬁlled.

The classiﬁer was trained on a randomly selected

data subset. The training subset contained 730 ERP

trials (described in detail in (Va

reka et al., 2014a))

with equal numbers of targets and non-targets. The

trained classiﬁer was subsequently applied to the data

of individual subjects.

The Matlab scripts available in (Va

reka et al.,

2014b) and using EEGLAB and BCILAB functions

were used for the implementation. Feature extrac-

tion follows the Windowed Means Method proposed

in (Blankertz et al., 2011). This method includes fea-

ture extraction: low pass ﬁltering and spatial ﬁltering,

and machine learning technique based on one of the

deep learning models - stacked autoencoders. Feature

extraction was described in detail in (Va

reka et al.,

2014a). Machine learning was designed as follows.

The Matlab implementation of stacked autoen-

coders was used. The parameters (including number

of layers, number of neurons in each layer, etc.) were

empirically optimized. The experimentation started

with two layers, then either new neurons were added

into the layer, or a new layer was added until the per-

formance of the classiﬁer stopped increasing.

Finally, the following procedure was used to train

the network:

1. The ﬁrst autoencoder with 100 hidden neurons

was trained. The maximum number of training

epochs was limited to 500.

2. The second autoencoder with 75 hidden neurons

was connected with the ﬁrst autoencoder to form

a 133-100-75-133 neural network, and trained.

The maximum number of training epochs was

limited to 300.

3. The third autoencoder with 60 hidden neurons

was connected with second ﬁrst autoencoder to

form a 133-100-75-60-133 neural network, and

trained. The maximum number of training epochs

was limited to 200.

4. The fourth autoencoder with 30 hidden neu-

rons was connected with third autoencoder to

form a 133-100-75-60-30-133 neural network,

and trained. The maximum number of training

epochs was limited to 200.

Furthermore, the following parameters were set

for the network globally: L2WeightRegularization

was set to 0.004, SparsityRegularization was set to 4,

and SparsityProportion was set to 0.18.

After the training of each autoencoder, the input

feature vectors were encoded using that autoencoder

to form input vectors of the next autoencoder.

Using the output of the last autoencoder, softmax

supervised classiﬁer was trained with 200 training it-

erations. Finally, the whole pre-trained 133-100-75-

60-30-2 network was ﬁne-tuned using backpropaga-

tion.

The learned model was ﬁrst veriﬁed on other

P300-based data (Va

reka et al., 2014a). Then, for

each subject, error rates depicted by red bars were ob-

tained by applying the classiﬁer in the testing mode.

Let us suppose that we have t

- number of correctly

classiﬁed targets, t

- number of correctly classiﬁed

non-targets, f

- number of misclassiﬁed non-targets,

- number of misclassiﬁed targets. The error rate

was calculated according to Equation 1.

ERR =

f p + f n

t p + tn + f p + f n

(1)

As a result, error rates indicate the extent to which

the classiﬁer was unable to separate target and non-

target single trials. The classiﬁcation results may

slightly differ with each run because of the indeter-

ministic training process.

6 DATA PREPROCESSING

The recorded EEG/ERP data as well as the data ob-

tained from the respiratory sensor were processed us-

ing the following workﬂow:

• Channel selection: The following channels cap-

turing brain data were selected for the initial pro-

cessing: Fp1, Fp2, F3, F4, C3, C4, P3, P4, O1,

O2, F7, F8, T3, Fz, Cz, and Pz.

• Driving session selection: Data for each driving

session were processed separately.

• Data ﬁltering: IIR Butterworth ﬁlter (frequency

range 0,01 Hz - 20 Hz) was applied to the data.

• Data segmentation: The epochs were extracted

from datasets, data corresponding to each target

and non-target stimulus were selected in the time

interval (-100 ms before the stimulus, 1000 ms af-

ter the stimulus) in the area of occurrence of the

target or non-target stimulus.

• Application of the ﬁlter for automatic artifacts de-

tection: The segmented data exceeding the range

(-100 microV, 100 microV) were denoted as pos-

sible artifacts and provided for manual inspection.

• Rejection of corrupted data: The data automati-

cally denoted as artifacts were manually inspected

Experimental Design and Collection of Brain and Respiratory Data for Detection of Driver’s Attention

445

0,00

5,00

10,00

15,00

20,00

25,00

30,00

35,00

40,00

45,00

50,00

0001

0003

0004

0005

0006

0007

0008

0009

0010

0011

0012

0013

0014

0015

0016

Error rate (%)

Participant

Error rates of the P300 component (%)

Figure 1: Results of validation. The error rates for each subject are depicted in bars. Higher error rates mean lower amplitudes

of P3s and/or more distortion in the EEG/ERP signal.

and most of them rejected. Moreover, each dataset

was manually inspected and in case of suspected

artifacts the related epochs were rejected. Arti-

facts were usually caused by eyes blinking, swal-

lowing or movements.

• Baseline correction: The baseline was corrected

using the interval (-100ms, 0ms) before occur-

rence of each target or non-target stimulus.

• Data averaging: The accepted epochs for each

participant and each session were averaged and

stored separately for target and non-target stimuli.

• The data captured by the respiratory sensor were

processed using the following workﬂow: only the

.vhdr ﬁle was read, ﬁltering was applied, and the

scale for respiration visualization was adjusted.

7 DATA PROCESSING

For the next analysis only the channels P3, P4, Fz,

Cz, and Pz (since the occurrence of the P3 compo-

nent is more signiﬁcant on these channels) were se-

lected. Then grand averages for each driving ses-

sion and each participant were computed (separately

for target and non-target stimuli). The latency of the

P3 component was determined using the technique of

peak latency. It is the simplest way to determine the

latency of the P3 component when the maximum am-

plitude in the time frame of possible occurrence of

the P3 component is searched for. The P3 component

time frame was set to (300 ms, 450 ms) reﬂecting the

expected location of peak latency values occurrence

in case of auditory stimulation. Finally, the peak la-

tency was determined manually from computed grand

averages.

8 RESULTS AND DISCUSSION

The results from performed experiments are summa-

rized in the ﬁgures and tables presented further. The

P3 component for the participant 0010 and the val-

ues of his/her peak latency for the ﬁrst driving session

and for channels P3, P4, Fz, Cz, and Pz are shown in

Figure 2.

It can be seen that the component P3 is clearly

identiﬁable and there is a substantial difference be-

tween reaction to target and non-target stimuli. How-

ever, the identiﬁcation of the P3 component was not

so evident for all participants. This is shown in Ta-

ble 1 where peak latencies for selected channels and

each driving session are available.

The data collected from participant 0002 were

rejected, the resulting values were out of reason-

able range, most probably because of technical fail-

ure during the measurement. The data from partici-

pants 0005, 0006, 0008, and 0009 were also not fur-

ther interpreted since the N2 component (a repetitive,

non-target stimulus elicits N2 deﬂection that can be

thought of as the basic N2 component) had a very high

amplitude while the component P3 was not clearly ev-

ident (an example for the participant 0006 is available

in Figure 3). It can also easily observed that the am-

plitude of the N2 component decreased in time; the

difference in peak amplitudes of this component is

HEALTHINF 2017 - 10th International Conference on Health Informatics

446

Figure 2: Grand average for participant 0010, the ﬁrst driving session, channels P3, P4, Fz, Cz, and Pz, peak latency of the P3

component for target and non-target stimuli.

noticeable comparing the waveforms for each driving

session. It was probably caused by habituation of par-

ticipants during the course of the experiment.

However, it does not mean that the data contain-

ing high N2 amplitude and low P3 amplitude are in-

correct. Since we were unsure with their correct inter-

pretation, they were rejected only for this preliminary

analysis. It is reasonably possible that they will be

considered for later analysis after reviewing them by

another expert in the ﬁeld. The data from participants

0012 and 0013 were not interpreted because of low

amplitude of the P3 component.

It is also evident that peak latencies differ among

the participants in average. This is a natural phe-

nomenon that reveals that the cognitive processing of

stimuli is different for each individual in the consid-

ered time frame. It can be also seen that the aver-

age latency for all participants for each selected chan-

nel increased as the experiment continued to the next

driving sessions.

However, different results we got by computing

grand averages for all epochs containing the target

stimulus and for all participants. It is not an average

of peak latencies as computed above, all waveforms

related to the target stimulus within driving sessions

were averaged. The resulting waveforms are given in

Figure 4. The determined values of peak latency for

averaged waveforms are clearly shown in Table 2. It

can be seen that grand averages for target stimuli, se-

lected channels and all participants differ between the

ﬁrst and second driving session, while there is no ev-

ident difference between the second and third driving

session.

The respiration rate was computed for each par-

ticipant and for each driving session as it can be seen

in Table 3. The average respiratory rate decreased in

time. Participants 0013 and 0014 had an increased

respiratory rate in all driving sessions in general, but

it also decreased within the course of the experiment.

The values from participants 0002 and 0015 were not

captured correctly due to technical difﬁculties with

the sensor.

9 CONCLUSIONS

This paper described the experiment dealing with at-

tention of drivers during a simulated drive. Brain ac-

tivity and respiratory rate of the participants of the

experiment were measured and investigated by us-

ing mainly the methods of electroencephalography

and event related potentials. Drivers, university stu-

dents, were stimulated by simple auditory stimuli

while driving a car simulator on a monotonous track.

The collected data were annotated, stored, prepro-

cessed, validated, and partly analyzed. The peak la-

tency of the P3 component was derived from the data

and grand averages for each participant and driving

session as well as grand averages for all participants

and each driving session were computed.

Experimental results showed that the P3 compo-

nent had been identiﬁed at most participants during

all driving sessions. However, some experimental re-

sults were not interpreted in this article because of the

high amplitude of the N2 component compared to the

amplitude of the P3 component. Prolongation of the

peak latency of the P3 component was evident in case

of most participants and in case of simple averaging of

Experimental Design and Collection of Brain and Respiratory Data for Detection of Driver’s Attention

447

Table 1: Grand averages for driving sessions, participants and selected channels.

Participant

1st driving session 2nd driving session 3rd driving session

Peak latency of P3 component [ms] on

channels

Peak latency of P3 component [ms] on

channels

Peak latency of P3 component [ms] on

channels

P3 P4 Fz Cz Pz P3 P4 Fz Cz Pz P3 P4 Fz Cz Pz

0001 423 421 336 410 418 452 450 404 406 486 497 507 402 400 497

0002 rejected because of technical failure

0003 383 384 382 386 383 390 416 394 399 378 398 431 387 398 346

0004 459 443 422 435 445 464 462 409 456 455 473 476 454 465 469

0005 too high amplitude of N200 component

0006 too high amplitude of N200 component

0007 345 334 323 343 341 356 353 357 356 354 357 355 359 358 356

0008 too high amplitude of N200 component

0009 too high amplitude of N200 component

0010 361 356 357 357 359 389 389 391 389 389 399 397 394 398 396

0011 317 318 347 305 322 319 311 309 303 327 363 410 307 304 407

0012 Low amplitude of P3 component

0013 Low amplitude of P3 component

0014 351 348 358 347 350 399 336 353 337 335 448 359 399 363 365

0015 444 413 404 406 411 428 418 411 415 418 498 505 448 456 449

0016 405 404 392 387 404 418 416 404 406 415 404 389 402 393 399

Avg 388 380 369 375 381 402 395 381 385 395 426 425 395 393 409

Figure 3: Grand average for the participant 0006, channels P3, P4, Fz, Cz, and Pz, peak latency of the N2 component for

target stimulus.

their peak latencies. Despite expectations, prolonga-

tion of peak latency in time was not clearly observed

when grand averages for all participants were inves-

tigated. This prolongation is evident only as a dif-

ference between grand averages of the ﬁrst and sec-

ond driving sessions. We also supposed that when the

driver expected his/her drive to be almost completed

(during the third driving session), his/her attention in-

creased. The average respiration rate and respiration

rates for most participants showed a decreasing trend

during the course of the experiment.

The results in this article were not statistically

evaluated. However, the trend of increasing latency

and decreasing respiratory rate is clearly visible. The

experimental results were naturally affected by differ-

ent brain reactions of participated drivers and sensi-

bility of captured data to the environmental noise, par-

ticipants’ overall mental conditions, and their move-

ments that caused occurrences of artifacts. Although

there was a big effort to eliminate these circumstances

by experimental design, setting of experimental con-

ditions, and usage of data preprocessing and process-

HEALTHINF 2017 - 10th International Conference on Health Informatics

448

Figure 4: Grand averages for all participants, channels P3, P4, Fz, Cz, and Pz, peak latency of the P3 component for the target

stimulus.

Table 2: Grand averages of peak latencies of the P3 component in [ms] for selected channels, target stimulus.

Grand average

1st driving session 2nd driving session 3rd driving session

P3 P4 Fz Cz Pz P3 P4 Fz Cz Pz P3 P4 Fz Cz Pz

391 391 369 386 389 407 407 395 402 405 406 405 395 402 402

Table 3: Respiration rate for each participant and driving

session.

Respiratory rate

Participant 1st driving ses-

sion

2nd driving

session

3rd driving

session

0001 18 16 16

0002 rejected

0003 16 15 14

0004 15 14 14

0005 19 18 18

0006 16 16 15

0007 16 15 15

0008 18 17 16

0009 17 17 15

0010 16 15 15

0011 14 13 13

0012 18 18 18

0013 23 22 22

0014 26 25 25

0015 rejected

0016 16 16 16

Avg 17,69 17,00 16,62

ing methods, they could not be completely removed.

That is why we also validated the collected data using

the stacked autoencoder.

The observations collected by experimenters from

the participants were also summarized. The partic-

ipants (they ﬁlled in the questionnaire) did not re-

port a highly increased level of fatigue after having

ﬁnished their experiment. Although the participants

complained about the tedious ride, the stimulation

kept them relatively attentive. One third of partici-

pants pointed out that they would like to drive longer

without any brake. On the other hand, most partic-

ipants complained about unpleasant feelings caused

by the ground electrode placed on their ear. This feed-

back have been used for the design of further ongoing

experiments.

We believe that by using the results of ongoing

research and technological innovations it will possi-

ble to capture biosignals more easily in the future.

It would facilitate recognition of the human attention

level and decrease the number of accidents not only

in transport but also during activities that are directly

inﬂuenced by human attention.

ACKNOWLEDGMENTS

This publication was supported by the UWB grant

SGS-2016-018 Data and Software Engineering for

Advanced Applications and the project LO1506 of the

Czech Ministry of Education, Youth and Sports under

the program NPU I.

Experimental Design and Collection of Brain and Respiratory Data for Detection of Driver’s Attention

449

REFERENCES

Blankertz, B., Lemm, S., Treder, M., Haufe, S., and M

uller,

K.-R. (2011). Single-trial analysis and classiﬁca-

tion of erp components – a tutorial. NeuroImage,

56(2):814 – 825. Multivariate Decoding and Brain

Reading.

Bo, O., Changxu, W., Guozhen, Z., and Jianhui, W. (2012).

P300 amplitude reﬂects individual differences of nav-

igation performance in a driving task. International

Journal of Industrial Ergonomics, 42(1):8–16.

Brain Products (2014). Brain vision recorder.

Bueno, M., Fabrigoule, C., Deleurence, P., Ndiaye, D., and

Fort, A. (2012). An electrophysiological study of the

impact of a forward collision warning system in a sim-

ulator driving task. Brain Research, 1470:69–79.

EEG/ERP Portal (2008-2016). EEG/ERP Portal.

Ichiki, M., Ai, G., and Wagatsuma, H. (2015). Simultane-

ous recording of eegs and eye-tracking for investigat-

ing situation awareness and working memory load in

distracted driving: A prospective analysis toward the

neuro-driving framework. Frontiers in Neuroscience,

(10).

Li, W., He, Q.-C., Fan, X.-M., and Fei, Z.-M. (2012). Eval-

uation of driver fatigue on two channels of eeg data.

Neuroscience Letters, 506(2):235–239.

Luck, S., Woodman, G., and Vogel, E. (2000). Event-related

potential studies of attention. Trends in Cognitive Sci-

ences, 4(11).

Luck, S. J. (2005). An Introduction to the Event-Related Po-

tential Technique (Cognitive Neuroscience). A Brad-

ford Book, 1 edition.

Mou

cek, R. and Ko

r, V. (2014). Attention of driver dur-

ing simulated drive. HEALTHINF 2014 - 7th Inter-

national Conference on Health Informatics, Proceed-

ings; Part of 7th International Joint Conference on

Biomedical Engineering Systems and Technologies,

BIOSTEC 2014, pages 543–550.

Mou

cek, R. and

richa, J. (2012). Driver’s attention dur-

ing monotonous driving. 2012 5th International Con-

ference on Biomedical Engineering and Informatics,

BMEI 2012, pages 486–490.

Mou

cek, R. and

Rond

ık, T. (2012). Inﬂuence of mental

load on driver’s attention. Transaction on Transport

Sciences, 5(1):21–26.

Neuro Behavioral Systems (2014). Home page.

Schier, M. (2000). Changes in eeg alpha power during simu-

lated driving: A demonstration. International Journal

of Psychophysiology, 37(2):155–162.

Strayer, D., Turrill, J., Cooper, J. M., Coleman, J. R.,

Medeiros-Ward, N., and F., B. (2015). Assessing cog-

nitive distraction in the automobile. Human Factors:

The Journal of the Human Factors and Ergonomics

Society, 57(8):1300–?1324.

SYNETIC GmbH (2014). World racing 2.

reka, L., Br

uha, P., and Mou

cek, R. (2014a). Event-

related potential datasets based on a three-stimulus

paradigm. GigaScience, 3(1):35.

reka, L., Br

uha, P., and Mou

cek, R. (2014b). P3-

validator.

Wester, A., B

ocker, K., Volkerts, E., Verster, J., and Ken-

emans, J. (2008). Event-related potentials and sec-

ondary task performance during simulated driving.

Accident Analysis and Prevention, 40(1):1–7.

HEALTHINF 2017 - 10th International Conference on Health Informatics

450