IEEE Consumer Electronics Magazine - October 2015 - 66

one of the most widely researched
fields in human-computer
interaction has been the use of
human eye gaze as an input
modality to control and command
intelligent systems.
devices and systems. No one wants to start looking like a Borg
to control the TV set!
More recent research has employed passive video-based gaze
trackers placed at a specific distance from the user, and gaze estimation is effected by capturing and processing images of the full
face or eye region in natural light or using active illumination at
near infrared (NIR) wavelengths. The NIR illumination is not
perceived by the subject, and such systems are nonintrusive.
Apart from improved user convenience, another advantage of
these passive systems is their low setup cost, as they just use one
or two cameras and a few LEDs [3].
Despite significant efforts by researchers in this field, the
usability of gaze cues in the consumer domain remains marginal in terms of accuracy and reliability, and there are additional
issues surrounding the suitability of eye gaze for calibrationfree use cases.
In this article, we discuss some existing applications that
use human eye gaze as a vital cue for various consumer platforms. For each of the use cases, the utility and advantages as
well as inherent limitations are discussed.

METHODS FOR DETERMINING
THE EYE GAZE POINTS

be broadly categorized into two general approaches:
1) appearance-model-based and 2) feature-based methods.
We give a brief description for each of these approaches in
the next sections.

AppeArAnce-model-bAsed methods
The appearance-model-based methods use the general shape of
the eyes and position of the pupils relative to the eye corners to
find the point of gaze [4]-[6]. A pretrained model of the shape
and appearance (texture and lighting) of the eye region is fitted
to a sequence of image frames [2], wherein it provides a fit to
eye regions depending on whether the model has been trained
with sufficient data that match the acquisition conditions and
physical characteristics of the input eye region(s).

AdvAntAges And disAdvAntAges of the
AppeArAnce-model-bAsed methods
The main advantage of these approaches is their low hardware requirements. Usually, methods in this category are
suitable for implementation on platforms without a high-resolution camera or additional light sources.
The main disadvantage of this class of methods is low accuracy, typically being 2 or 3˚ for fixed head position [4], [5], [7]. As a
result, it can be stated that small changes in eye gaze (1-2˚ difference), which can be important for smartphone applications, may
not be detected by the appearance-based methods (depending on
size of the smartphone, the angle spanned by the pupil can be very
small-7-10˚). Other problems arise with changes in head pose,
variation in global illumination due to different directional light
sources, facial expressions, and skin color. Compensating for all
of these issues requires large training data sets and makes appearance-based methods computationally expensive.

FeAture-bAsed methods

In the following sections, we cover some of the most interesting approaches developed for video-based gaze trackers
and their implementation. Eye-gaze estimation methods can

These methods take into account various characteristics of
the human eye to identify a set of distinctive features like
contours (limbus and pupil contour), eye corners, and cornea
reflections of NIR illuminators (LEDs). These approaches

Right Viewing Window
of HMD
Center of
Cornea Sphere
Center of Pupil

Gaze Point
Gaze Vect
r
or
ito
on
M
l
X
a
Illumination
tu
Vir
IR-LED

Cornea Sphere

Camera
Collimated IR-LEDs

(a)

(b)

FIGURE 1. (a) A schematic for a feature-based eye-gaze-tracking method [27]. (b) tracking nir reflections on the cornea and pupil center in the pccr method [17].
66 IEEE Consumer Electronics Magazine

october 2015

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