22-04-2014, 02:34 PM
Pupil/EEG-based Drowsiness Detector
ABSTRACT
In the following report, we present our topic for the Final Year Project: A Pupil/EEG-
based Drowsiness Detector. After presenting our purpose behind the selection of this
topic, we present a literature review about what has been done so far in the field.
Thus, we will be building and testing a Drowsiness Detector for use in transportation,
security and control jobs, and other kinds of situations that require a high level of
alertness.
Since many of the accidents occur because of inattention or sleepiness, our design will be
used to monitor fatigue and produce an alarming signal in order to alert the individual.
Unlike many models built so far, which use single detecting techniques, such as the study
of electroencephalographic (EEG) signals, the eye parameters (such as blink rate,
duration and frequency), the PERcentage of eye CLOSure (PERCLOS), or
electrooculograms (EOG), our design will base its results on combining two techniques:
The study of the alpha activity in the EEG signal and the eye behavior, which proved to
be two of the most accurate and widely used techniques in the study of drowsiness.
Experiments were performed on many subjects in a controlled environment.
The coherence between the two results will be studied and an alarming signal will be
produced depending on the seriousness of the fatigue state.
Introduction
Sleep is an essential component of daily life: It helps keep the mind and body at rest,
consolidates
memory,
and
promotes
learning,
concentration
and
accuracy...
The human brain is only capable of sixteen hours of wakefulness beyond which it cannot
perform assigned tasks efficiently. “Sleepiness can be defined as a variable, relatively
extended, and actively regulated reoccurring process that is influenced by levels of
arousal and sleep need.” It is manifested twelve hours after mid-sleep and before
nighttime sleep and it is related to lifestyle issues but becomes problematic in some
situations [1].
People try to overcome their sleep deprivation by caffeine intake, a few exercises, or
other activities. However, these techniques work only temporarily. The individual soon
finds himself nodding behind the wheel, in a meeting, or in a monotonous task in a
nuclear power plant. High levels of alertness are important for people in some situations,
where their life can be at risk if they are inattentive or fall asleep.
This is why a model that monitors fatigue and drowsiness is useful in transportation,
control environments, security, meetings, and others.
Methods for alertness monitoring have been proposed and some models were
implemented.
Drowsiness Detection Techniques
General measurements
Techniques such as increased reaction time to any occurring event, grip strength (in the
case of drivers), skin conductance and decreased production of the hormones adrenaline,
noradrenalin and cortisol can be used in order to detect signs of drowsiness.
Moreover, heart rate can be observed experimentally by using special kinds of sensors in
order to monitor alertness.
However, these techniques are inaccurate because they can be indicative of physiological
states other than drowsiness. This is why it is better to use them with more accurate
techniques in order to validate the resulting data.
Reaction time measurement
Since it is relatively difficult to remain alert when performing monotonous tasks (such as
listening to long lectures, observing activities in a mine, driving at early hours in the
morning...), short stimuli can be applied to bring the subject back to the alert state in a
few minutes. An auditory or visual test can be performed where the stimulus is presented
randomly (based on a random counter value). The subject reacts by pressing a button and
the time difference between the stimulus and the response can be calculated to evaluate
the reaction time of the subject. The analysis of the variation in reaction time for the same
person in defined experimental conditions is indicative of the level of drowsiness.
System Design
Based on the information available about various techniques to detect drowsiness, it was
found that the best technique is the monitoring of the eyes’ behavior along with the
analysis of EEG signals acquired from the brain.
Other techniques are complex, costly and hard to implement, or are imprecise and
inaccurate. The choice of the brain and the eyes as the focus of our study can be
explained if the structure and function of both organs can be elicitated.
EEG signal amplifier
The analysis of the EEG signal should be done with high accuracy. Since the EEG signals
have a very low voltage (in the order of few micro volts), we need to amplify the signals
to have them ready for signal processing. Open EEG offers a modular solution where
amplification is done without aliasing and signal distortion. Open EEG is an open source
website where many hardware implementations that help study EEG signals are offered.
The modular EEG is a hardware solution where the acquisition and amplification of EEG
signals is provided. The amplified signals are tested and analyzed for various
applications.
The board is composed of many filters and amplifiers that have various properties. The
filters provide the functions needed to make a robust hardware and ensure safety and
efficiency under normal circumstances.
In addition to the amplifier board, an A/D converter is needed in order to digitize the
amplified signal for subsequent computer analysis. The A/D converter tool in LabView
will be used since building the converter will add cost and consume effort and is outside
the scope of our project.
Eye detection algorithm
Since the light is reflected in different ways from the eyes and from the face and since the
“eye region presents great intensity changes, the eyes are located by finding the
significant intensity changes in the face” [16].
Image acquisition is the first step in the design and it is done using a webcam. The
webcam was chosen for its relatively low cost, small size and suitable number of frames
grabbed per second, which satisfies the simplicity, practicality and efficiency constraints
of our model. The eye detection and localization algorithm is presented in Appendix A.