

Drowsiness Detection by Electrooculogram Signal Analysis in Driving

Simulator Conditions for Gold Standard Signal Generation

N. Rodríguez-Ibáñez

, P. Meca-Calderón

, M. A. García-González

, J. Ramos-Castro

and M. Fernández-Chimeno

Biomedical Engineer at Ficosa International S.A, Can Magarola,

Ctra. C-17 km 13. E-08100 Mollet del Valles, Barcelona, Spain

Software Engineer at Idneo Technologies S.L, Pol. Ind. Can Mitjans, 08232, Viladecavalls, Barcelona, Spain

Group of Biomedical and Electronic Instrumentation of the Department of Electronic Engineering of the Technical

University of Catalonia (UPC), Barcelona, 08034, Spain

Keywords: Drowsiness Detection, Gold Standard, Driver Monitoring, Electrooculogram, Electroencephalogram.

Abstract: Detection of drowsiness while driving is a leading objective in advanced driver assistance systems. This

work presents a new index to assess the alertness state of drivers based on the EOG dynamics derived from

a polysomnography device. More than 15 hours of laboratory tests were analyzed in order to detect

drowsiness while doing cognitive activities. The proposed method has a sensitivity of 92, 41% and a VPP of

93,41% in detecting drowsiness. The results show that the proposed index may be promising to assess the

alertness state of real drivers.

1 INTRODUCTION

Drowsiness is one of the main causes of vehicle

accidents. A recent study showed that 20% of

crashes and 12% of near-crashes were caused by

drowsy drivers (NHTSA VSR, 2006). The morbidity

and mortality associated with drowsy-driving

crashes are high, perhaps because of the higher

speeds involved combined with delayed reaction

time (Faber, 2004).

Driver behaviour monitoring, and the reliable

detection of drowsiness and fatigue is one of the

leading objectives in the development of new

Advanced Driver Assistance Systems (ADAS). Of

the use of biomedical signal analysis to detect

drowsiness in real vehicles appears the need of an

objective gold standard to compare with the selected

signals, in this case thoracic effort. The most

objective signal to assess the sleep onset phase is

Electroencephalography (EEG). The problem

associated to this signal is that, in real environments

(i.e. vehicles) the actual devices used in hospital

environment to acquire de data presents artefacts due

to vibration and movements of the vehicle that

masks the real EEG signal.

The aim of this work is validate the EOG signal

as a new Gold Standard and the EOG acquisition

device as a good quality device to ensure the optimal

quality of the data. The EOG signal is a highly

robust to artifacts signal related to EEG valuable to

compare with our drowsiness detection index based

on thoracic effort variability (TEDD) in real

environments. (Rodríguez-Ibáñez, 2011)

2 PRIOR WORK

2.1 EEG and EOG Signals as Gold

Standard

During active wakefulness (i.e., when the person is

awake and pursuing normal activities), the EEG is

characterized by high frequencies (i.e., 16 to 25 Hz)

and low voltage (i.e., 10 to 30 microvolts). EOG

readings during wakefulness exhibit Rapid Eye

movements (REM).

During relaxed wakefulness (i.e., when a person

is awake but has his or her eyes closed and is

relaxed), the EEG is characterized by a pattern of

alpha waves with a frequency of 8 to 12 Hz and an

amplitude of 20 to 40 microvolts. EOG readings

Rodríguez-Ibáñez N., Meca-Calderón P., García-González M., Ramos-Castro J. and Fernández-Chimeno M..

Drowsiness Detection by Electrooculogram Signal Analysis in Driving Simulator Conditions for Gold Standard Signal Generation.

DOI: 10.5220/0004241500570063

In Proceedings of the International Conference on Biomedical Electronics and Devices (BIODEVICES-2013), pages 57-63

ISBN: 978-989-8565-34-1

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

show slow, rolling movements (Roehrs, 2011),

increase of blinking frequency and lots of saccadic

response at the transition to NREM sleep onset

(Dinges, 2005).

2.2 EEG and EOG Signals Acquisition

in Real Driving Environments

The most important handicap in the field of

drowsiness detection in real driving environments is

the fact that the filtration of the low amplitude

biomedical signals in order to eliminate vibration

and movement artifacts is a very complex work that,

in most cases, also affects the original signal of

interest.

Hundreds of real vehicle tests have been made in

the last three years with the objective of finding a

biomedical signal robust to artifacts and also related

to sympathetic-vagal system to provide drowsiness

information in real vehicle tests.

The EEG signal has always been the most

objective signal to define drowsiness in laboratory

conditions but in real vehicle tests the EEG signal

presents several problems as artefacts and the fact

that the EEG codifications Rechtschaffen & Kale’s

method (Rechtschaffen & Kale, 1968) is only

recommended with closed eyes. According to the

EEG-EOG studies there is a relation between EEG

waves and EOG patterns that allows generating an

objective Gold Standard signal for drowsiness

detection from EOG signal.

For the first real vehicle tests the EEG and EOG

signal was acquired with a Bitmed eXim Pro

polysomnography device. The EOG signal quality

was good before and after filtering the vehicle

vibrations and movement artifacts but the EEG

signal was lost in the filtering process due to the fact

that the frequency of the vibrations was the same

frequency that the waves of interest (theta and alpha

waves).

Following this results, currently we have focused

on find new devices that avoids the problem of the

artifacts in EEG signal. Two different tests have

been made in real vehicle with two different

polysomnography devices:

- Nicoletta wireless device

- Bionic EEG holter that provides active electrode

technology

Although both systems show improvements in the

EEG signal quality it hasn’t enough quality to

extract the drowsiness information. The filtering

solution had the same problems that with other

polysomnography devices.

Taking in to account this results and the fact that

the EEG and the EOG signals are physiologically

related we recommend the use of EOG data as Gold

Standard in real vehicle tests. This work proposes

different indexes based on slow eye movement’s

detection, blinking frequency and saccade

movement’s inhibition.

3 MATERIALS AND METHODS

3.1 Measurement Protocol

The participants in the test were 17 male and 6

female with ages between 20 and 29 years and no

clinical conditions. These tests were designed and

performed in laboratory conditions.

To perform these tests the setup was equipped

with a biomedical monitor (Bitmed eXim Pro,

BitMed) and a webcam. The biomedical signals

selected as significant for this test were the external

observer (video), Electrooculography (EOG) and

thoracic effort. The thoracic effort signal was

measured in all cases using an inductive band

located at the middle trunk above the diaphragm.

The EOG signal was measured with four

Electromyography (EMG) single electrodes: two

were located in the outer cantus of each eye in the

case of the horizontal EOG setup, and two more

electrodes located in the upper part and in the lower

part of the right eye (Fig. 1). The EOG and the

respiratory signal were sampled at 100 Hz.

Figure 1: EOG instrumentation.

Video signal was recorded to generate the

external observer variable.

3.2 Test Design

The test was designed to classify the different eye

movements and set a level of eye activity or eye

inactivity (related to drowsiness). The test setup

consists of a vehicle seat and a 19’’ inches monitor

in front so the subject of the tests can see the

patterns classification video seated on the vehicle

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

seat. The test has two parts:

a) Patterns classification part.

Once the subjects are seated and connected to the

acquisition systems the first part starts and they were

asked to watch a 5 minutes video with the objective

of follow the point on the video movements of the

point in the screen represents the movement of the

eye for the following patterns of interest:

- Saccadic movement

- Compensation movements

- Blinking

- Fixed gaze

- Seeking movements

- Slow Eye Movements (SEM)

The monitor has to be no more than 15 cm far from

the face of the subject.

b) Drowsiness state classification part

The subject rest relaxed in the seat for over 20

minutes with eyes open.

3.3 Patterns Classification

The patterns selected as indicative of drowsiness

where the following:

3.3.1 Saccadic Movement

Saccadic movements are defined as rapid symmetric

eye movements with the objective of constantly

change the retinal focus from one point to the next

point in the visual path.

There is a linear relation between the size of the

saccade and the velocity of the ocular movement.

The mean duration of saccadic movements ranges

between 30 and 120 ms.

In an awaken state these movements are mostly

voluntary and they are used to redirect the gaze to

the point of interest of the scene. In fatigue and

drowsy states the saccadic speed decreases (Galley,

1989, 1993, 1998; Sirevaag & Stern, 2000) and the

latent period between saccades increases.

3.3.2 Compensation Movement

Compensation movements are reflex movements

that imply the coordination of both eyes. These

movement works as an object fixation mechanism

while moving head or body. The most important is

the Vestibulo-Ocular Reflex (VOR) with a response

time of 16ms.

Figure 2: EOGv saccades (black) and EOGh saccades in

red. Filter: band-pass 0.2-30Hz.

Figure 3: Compensation movement in EOGv signal.

Band-pass filter 0.2-30Hz.

3.3.3 Blinking

Figure 4: Blinking pattern on EOGv signal. Band-pass

filtering 0.2-30Hz.

DrowsinessDetectionbyElectrooculogramSignalAnalysisinDrivingSimulatorConditionsforGoldStandardSignal

Generation

Blinking is the rapid closing and opening of the

eyelid that provides moisture to the eye by irrigation

using tears and a lubricant that the eyes secrete. The

mean frequency of blinks in a normal subject is 12

to 19 blinks per minute. This frequency can be

influenced by internal or external factors. Fatigue

and drowsiness decreases the blinking rate and

increases the percentage of eye closure time.

3.3.4 Fixed Gaze

The fixed gaze or ocular movement fixation can be a

characteristic pattern of interest in one point or low

cognitive activity depending on the duration of the

pattern. In a normal context the fixed gaze duration

ranges between 200ms and 350ms with open eyes.

In phases of fatigue or drowsiness the fixed gaze

time can reach 3 seconds becoming an ocular lost of

activity (Salthouse and Ellis, 1980); (Viviani, 1990).

Figure 5: EOGh fixed gaze pattern. Ban-pass filter 0.2-

30Hz.

3.3.5 Seeking Movements

Seeking movements are coordinated movements

between two eyes with the porpoise of follow slow

visual stimuli. Their function is to stabilize the

dynamic visual image in the retina with velocities

between 1 and 30º/s.

Figure 6: Seeking movement in EOGv signal. Band-pass

filter 0.2-30Hz.

3.3.6 Slow Eye Movements (SEM)

Slow eye movements are eye movements with

duration between 1 and 3 seconds mostly detected in

the horizontal component of the EOG. This

movement is characteristic of drowsiness states. Its

characteristic of sleep onset with eyes closed but this

pattern can also be seen with open eyes in drowsy

drivers fighting for not to fall sleep.

Figure 7: SEM. Band-pass filtering 0.2-30Hz.

3.4 Drowsiness Indicators

Awaken estate has been defined as a state of high

activity and information interchange between the

subject and the environment (Phase 0), Fatigue as a

state of lack of energy and motivation (Phase 1) and

drowsiness as a state related to the sleep onset. Only

some of the EOG patterns explained have direct

relation with the sleep onset:

Blinking – An increase of the blinking frequency in

addition to an increase of the percentage of eye

closure are indicative of sleep onset.

Saccade – The number of saccades and the detection

of fixations combined can be an index to estimate

the ocular activity assuming saccades as activity and

fixation as no activity. There is a direct relation

between the reaction time of the subject and the

velocity of the saccade movement.

Slow Eye Movements (SEM) – During the transition

of awake to sleep is very common the appearance of

slow eye movements (SEM), like pendulum low

frequency (0.1-1hz) movements in the horizontal

line of the eye.

Figure 8: Determination of the beginning and the final of

the saccade movement. Binocular motor coordination

during saccades and fixations while reading: A magnitude

and time analysis (Vernet et al., 2011).

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

4 EOG SIGNAL PROCESSING

4.1 Preprocessing

A non-linear filter preprocessing of the signal has

been done. The filter used was a non linear filter

derived from the Hodrick-Prescott (1) filter with the

objective of removing repeated oscillations in the

signal. Cutoff frequency at -6dB.

(1)

In secondly a band pass filter has been done. The

high pass filter at 0.1 Hz filtered the baseline

eliminating the electrode polarization effects and the

movement artifacts. The low pass filter at 30Hz

eliminated the Electromyogram (EMG) artifacts of

the signal.

4.2 Processing

As seen in the literature, the most representative

EOG patterns used to estimate the sleep onset are

saccade, blinks and slow eye movements. This

investigation was focused on the analysis of blinking

and saccade patterns as explained below.

4.2.1 Blinking Detectors

The analysis was divided in two blocks (Fig.9):

erosion and detection.

Figure 9: Block diagram of the blinking detection

algorithm.

First the signal passes the erosion block, where

the abrupt swings are eliminated (Fig.10). Then the

filtered signal passes to the blinking detection

module.

The objective of the erosion module is to stand

out the blinking patterns from the rest of artifacts

and saccade oscillations with the interpolation of the

obtained “yRET” signal and its posteriors

calculation of the very low frequency oscillations

obtaining “FPA 1Hz” signal. Finally the “FPA 1Hz”

signal is subtracted from the “yRET” signal to

obtain C signal.

Figure 10: Erosion block.

Peak detection

Figure 11: Detection block.

In the detection block C signal is processed with

the objective of stand out the low frequency

oscillations to avoid remaining artifacts. Finally the

subtraction of yRET signal from C is done and the

detection of peaks with a fixed threshold ‘Um’ gives

the resultant signal with the blinks detected.

4.2.2 Saccade Detectors

The saccade detection algorithm developed analyzes

the horizontal EOG signal with an adaptation of the

known Murty-Rangaraj method based on the

detection of QRS segment in EKG signal (Rangaraj,

2002).

As shown in the picture below (Fig. 12) the

analysis is divided in three blocks: The

preprocessing block explained in E.1, The Murty-

Rangaraj adaptation block and the saccade detection

block.

Figure 12: Block diagram of the saccade detection

algorithm.

Preprocessing

Murty Rangaraj adaptation

Quadratic

Derivate

MA,

M samples

Detection

DrowsinessDetectionbyElectrooculogramSignalAnalysisinDrivingSimulatorConditionsforGoldStandardSignal

Generation

Murthy-Rangaraj method consists in a pre-

filtering of the signal followed by and estimation of

the first weighted quadratic derivate (2). The

resulting signal was later filtered with a moving

average filter (3) in order to smooth the obtained

result.

(2)

(3)

Next step was the maximum and minimum

identification of the signal in order to detect the

position of the saccade using a fixed threshold.

5 STATISTICAL ANALYSIS

For each minute of recording, the phases obtained by

the EOG different drowsiness detection algorithms

were compared the GS signal, in this case a

combination of three external observers evaluating

minute by minute the state of the subject using a

video recording of the tests. To estimate the

sensitivity and specificity of the different EOG

methods a match signal was calculated having the

number of false positives, false negatives, true

positives and true negatives.

According to Table 1 (Stone EA, 2005),

sensitivity (Sens) and specificity (Spec) for each

phase is defined as:

Table 1: Sensitivity and Specificity definition.

Gold Standard

PHASE 0 PHASE 2

EOG index

PH0

TN FN

PH2

FP TP

being the Sensitivity the proportion of actual

positives which are correctly identified as such

giving information about how good is the detection

algorithm, and the Specificity the proportion of

negatives which are correctly identified.

6 RESULTS

The results for the analysis of the EOG signal with

de blinking detection algorithm shows positive

results with a sensitivity of 92,41% and a VPP of

93,41% (Figure 14) comparing the results of the

algorithms with the Gold Standard. The results with

the saccade detection algorithms shows also good

results but, in this case, it has to be improved with a

module that allows the detection of the beginning

and the final of the saccade pattern in order to

improve the pattern detection, yet the results are

very promising for drowsiness detection porpoises

with sensitivity values of 85,1% and VPP values of

95,4% (Figure 15).

Figure 13: Example of the saccade detection in horizontal

EOG.

Figure 14: Blinking detection algorithm results.

Figure 15: Saccade detection algorithm results.

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

7 CONCLUSIONS

The results confirmed the viability of the sleep onset

detection using related to drowsiness patterns in the

EOG signal as blinking frequency and saccade

movements’ appearance. Some misdetection of the

algorithms may be due to the inter-subject variability

mostly regarding the shape of the saccade pattern.

Future work will be focused in the improvement

of the saccade detection algorithm by including the

detection of initiation and end of the saccade pattern

in order to make more specific the detection and

accurate the calculation of the variable velocity of

the saccade.

The future objective is to use the EOG signal as

Gold Standard in vehicle tests replacing the EEG

signal that shows low quality signal in real

environments.

ACKNOWLEDGEMENTS

This work has been partially funded by the Spanish

MINISTERIO DE CIENCIA E INNOVACIÓN.

Proyecto IPT-2011-0833-900000.Healthy Life style

and Drowsiness Prevention-HEALING DROP.

REFERENCES

P. Meca Calderón, (Jun 2011). “Estudio de la viabilidad de

la señal de EOG como indicador de somnolencia”.

PFE. Master Oficial en Enginyeria Biomèdica.

Department of Electronic Engineering of the

Technical University of Catalonia (UPC)

NHTSA VSR, (2006) "Chapter 3: objective 2, what are

the environmental conditions associated with driver

choice of engagement in secondary tasks or driving

while drowsy?

Faber, J., (2004) “Detection of different levels of vigilance

by EEG pseudospectra”, Neural Network World vol.

14, no. 3-4, pp. 285–290.

Rodríguez-Ibáñez N., García-González M. A., Fernández-

Chimeno M., Ramos-Castro J., (2011). “Drowsiness

detection by thoracic effort signal analysis in real

driving environments” Conf Proc IEEE Eng Med Biol

Soc., 2011:6055-8.

Timothy Roehrs, Ph.D., and Thomas Roth, Ph.D., (Feb 25,

2011). "Sleep, Sleepiness, and Alcohol Use". National

Institute on Alcohol Abuse and Alcoholism of the

National Institutes of Health”

Dinges D. F., Mallis M. M., Maislin G., Powell J. W.,

(2005) Evaluation of techniques for ocular

measurement as an index of fatigue and the basis for

alertness management.

Rechtschaffen A., Kales A., (1968). A manual of

standardized terminology, technique and scoring

system for sleep stages of human subjects los Angeles:

Brain information services/Brain research institute,

UCLA; 1-12.

Galley, N., (1989). Saccadic eye movement velocity as an

indicator of (de)activation. A review and some

speculations. Journal of Psychophysiology, 3, 229-

244.

Galley, N. (1993). The evaluation of the electrooculogram

as a psychophysiological measuring instrument in the

driver study of driver behaviour. Ergonomics, 36,

1063-1070.

Galley, N., (1998). An enquiry into the relationship

between activation and performance using saccadic

eye movement parameters. Ergonomics, 40, 698-720.

Rangaraj R., (2002) ‘Biomedical Signal Analysis, A Case

Study approach’, IEEE Press Series in Biomedical

Engineering.

Salthouse and Cl Ellis, (1980), Determinants of eye-

fixation duration. Am J Psychol. Jun;93(2):207-34.

Sirevaag, E. J. & Stern, J. A., (2000). Ocular measures of

fatigue and cognitive factors. In R. W. Backs & W.

Boucsein (Eds.), Engineering Psychophysiology (pp.

269-288). Mahwah, NJ: Lawrence Erlbaum

Associates.

Stone E. A, Cooper G. M., Sidow A., 2005.Trade-offs in

detecting evolutionarily constrained sequence by

comparative genomics. Annu. Rev. Genomics Hum.

Genet. 6:143–64.

Vernet et all. (2011) Guiding Binocular Saccades during

Reading: A TMS Study of the PPC, Front Hum

Neurosci. 2011; 5: 14. Published online 2011 February

7. doi: 10.3389/fnhum. 2011.00014.

Viviani P., (1990). Eye movements in visual search:

Cognitive, perceptual and motor control aspects, in

Eye movements and their role in visual and cognitive

processes. Kowler E. Editor. Elsevier Science,

Amsterdam. p. 353-394.

DrowsinessDetectionbyElectrooculogramSignalAnalysisinDrivingSimulatorConditionsforGoldStandardSignal

Generation