IEEE Consumer Electronics Magazine - October 2015 - 113

Table 4. Different features used by organizations trying to solve DFD.

(continued from page 112)

Lane
Deviation

SteeringWheel
Movement

Head
Mental
Movement Reaction
Skin
Eye
Detection Time
HRV Conductance EEG Features

Company

Product

Previous
Data

Scania [41],
[42], [52]

N/A

-

x

x

-

-

-

-

-

-

Volkswagen [53]

N/A

-

-

-

-

-

-

-

-

x

Volvo [52]

N/A

-

-

-

x

-

-

-

-

x

CMU
[28], [55]

N/A

-

-

-

x

-

-

-

-

-

U.S. Army [17]

N/A

-

-

-

-

-

-

-

x

-

NHTSA [5], [6],
[10], [48]

N/A

-

x

-

x

-

-

-

x

x

some systems use more than one feature to detect the drowsiness level.
Eye features provide various advantages for a DFD system
based on machine-vision architectures. The vision systems
with better time response were the ones that analyzed the driver's physiological features, as they could predict when the driver was falling asleep instead of verifying when the driver was
already asleep. Eye features (i.e., PERCLOS) were defined as
the most widely used option [8], [22], [29], [40], [49] and considered as a good measurement by some NHTSA reports [2],
[5], [6], [48].

EYE FEATUREs
There are several ocular measurements taken as physiological
estimations of the mental state of the driver. This group of evaluations is "considered to be the most suitable for online monitoring" [29] as it is nonintrusive and provides various solutions:
blink rate, percentage/average of eyelid closure (PERCLOS/
AVECLOS), gaze or eye movement, and pupil response.
Blink rate is directly affected by fatigue, eye injury, medication, and disease [63]. Fatigued eyes are more susceptible to
high blinking rates. "Eye blinks can be influenced by psychological and behavioral variables and provide information about
perceptual and cognitive activity" [49]. There are several examples of DFD components using this feature, and, most of the
time, the developers of the DFD system use more than one
method: Alertness Monitor (MTI Research, Inc.) [49], Blinkometer (IM Systems, Inc.) [49], Optalert (Sleep Diagnostics,
Ltd.) [8], and Insight (SMI) [8], among others.
The pupil response could be associated with the HRV [64]
as the pupil is regulated by the autonomous nervous system.
This response could give information regarding the mental
workload of the subject, depending on the dilations and constrictions of the pupil [64], [65]. EyeCheck (MCJ) is a product that measures the pupil reaction to light flashes, obtaining
the pupil diameter [8].
Eye tracking, eye movement, and gaze orientation basically refer to the same parameter. They identify the direction in

Drivers are not typically conscious
of how their capabilities could be
diminished due to drowsiness.
which the driver is looking. These data would allow the DFD
system to know if the user is looking forward. Some examples of systems using eye tracking are Smart Eye [40],
Optalert (Sleep Diagnostics, Ltd.) [8], Insight (SMI) [8],
Driver Fatigue Monitor (Attention Technologies) [8], and
FaceLab (Seeing Machines) [8], [57].
PERCLOS and AVECLOS are eyelid closure metrics that
are usually combined with the other eye features, depending
on the application. Drivers affected by fatigue or drowsiness
usually lose their facial muscle tone and have slower eyelid
movements and longer eyelid closures or microsleeps. Therefore, PERCLOS is defined "as the proportion of time in a
minute that the eyes are at 80% closed" [61]. This term is
also described as PERCLOS-80 [65]. Some other applications use the term PERCLOS-70, which applies to the proportion of time in a minute that the eyes are closed at least
70% [8]. AVECLOS is a less complex measurement than
PERCLOS: "the percentage of time the eyes are estimated to
be fully closed over a 1-min interval" [66]. Also, there are
several products using PERCLOS and/or AVECLOS; most of
them are used in combination with other features: Co-Pilot
(Richard Grace) [22], [57]; Optalert (Sleep Diagnostics, Ltd.)
[8]; Insight (SMI) [8], [66]; Driver Fatigue Monitor (Attention Technologies) [8], [51]; FaceLab (Seeing Machines) [8],
[51]; and Driver State Monitor (Delphi) [8], [39], [66].
Drowsiness detection and even prediction is possible by monitoring multiple eyelid movement features, and this work is
reported in [67] and [68]. More recent work involving eyelid
closure and HRV is reported in [69].
The DFD systems based on eye features involve the following stages: facial extraction methods, face detection and
october 2015

^

IEEE Consumer Electronics Magazine

113
Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - October 2015
