IEEE Consumer Electronics Magazine - October 2015 - 67

Table 1. A summary of hardware setup requirements and accuracies achieved by various
gaze-tracking methods in the literature.
Method

Required Setup

Accuracy

Appearance
model [13]

Webcam (720 × 576)
chinrest, distance: 75 cm

2−5°

Appearance
model [2]

vgA (640 × 480) distance:
40-70 cm

4°

Appearance
model [4]

Webcam (1,280 × 1024)
50-60 cm

1° fixed head
2° slight
motion

feature (eye-mod- one camera (8 mp), 25-30
el) based [14]
leds

1°

feature (model)
based [15]

two cameras, two rings of
leds for each camera

1°

feature (regression) based [16]

camera (1,024 × 768)
one nir source, 71 cm

0.4° head
fixed,
1° slight
head pose

can further be subdivided into two main branches: regressionbased and model-based methods [8].

regression-bAsed methods
In these methods (also called pupil center corneal reflection
[PCCR] techniques), the vector between the corneal reflections and
pupil center is tracked and mapped geometrically with a polynomial regression function to gaze coordinates on screen [Figure 1(a)].

model-bAsed methods

the prime hindrances to such
gaze-based password-entry systems
are the possibilities of identity theft
and that unintentional eye and
body movements may lead to
erroneous gaze points and
incorrect password entry.
shown in Table 1. The main disadvantage is the hardware requirements for implementing these techniques as they need several
light sources or multiple cameras. The other problem is that if the
NIR glints in the cornea disappear due to head movements or
additional glints appear in the eye from other NIR sources (e.g.,
halogen lights, sunlight, and reflections from eyeglasses), the
method will fail.

cAlibrAtion in eye-gAze estimAtion
Since the eye-gaze systems should detect the human's optical
axis but the actual point of gaze is defined by human visual axis
and the difference between these two axes is specific for each
person (it can be as high as 5˚ horizontally and 1.5˚ vertically
[12]), the gaze-tracking system should be calibrated for each
individual as an integral part of the tracking procedure. Calibration is done by looking at certain predetermined points and is
often a cumbersome process. The error in the calibration procedure is added to other sources of error, which usually leads to
significant overall tracking error.

The eye-model-based techniques use the geometrical model
of the human eye [Figure 1(b)] along with NIR light sources,
multiple cameras, and one or more mirrors to develop accurate ray tracing to estimate gaze direction [9]-[11].

APPLICATIONS OF EYE GAZE IN
CONSUMER ELECTRONIC DEVICES

pros And cons of feAture-bAsed methods

Reports from agencies like the U.S. National Highway Traffic
Safety Administration stating that the majority of fatal auto
accidents are due to driver inattention have sparked interest in

The advantages of feature-based methods include their higher
accuracy in comparison to the appearance-based methods, as

A Narrow Bandpass Filter
Attached in Front of the Camera

NIR Illuminators
(a)

eye-gAze trAcking in intelligent
driving AssistAnce systems

Camera
Capturing
D
Driver Face

Face Detection
Within Video

Locating Eye
Region

Pupil Detection
and Eye Tracking
Determination of
Alertness Level

Alarm
(b)

FIGURE 2. (a) A gaze-tracking setup in a car [26]. (b) A schematic of an appearance-based method for gaze tracking in a driver alert system.

october 2015

^

IEEE Consumer Electronics Magazine

67
Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - October 2015
