Page 381 -

13.2 Eye tracking 371

new viewpoint (Duchowski, 2007)—perhaps in anticipation of a new task or in re-
sponse to some stimulus. These transitions lead to fixation—focus on a new area of
interest. However, fixation does not mean lack of motion—even when focused on a
target; eyes will continue to move in small microsaccades, which are essentially ran-
dom noise (Duchowski, 2007). Following a moving target (as in a video game) leads
to a final class of eye movements known as smooth pursuits.
Sophisticated software uses the geometry of the eye and the related optics to
filter out the noise and to identify saccades and fixations, providing highly accurate
measures of where the user is looking at any given time. The first step in this process
is generally to remove noise, often by ignoring measurements that are not plausible
given the operating characteristics of the eye tracker. De-noised movements are then
separated into saccades and fixations through one of two approaches. Dwell-time
methods look for periods of little or no variance in eye position. Low-variance inter-
vals lasting for more than some minimal amount of time are classified as fixations,
with other intervals classified as saccades. Velocity-based methods take the opposite
approach, classifying saccades as intervals when eye-movement velocity exceeds a
given threshold. Experience from prior literature can be used to select appropriate
parameters for fixation intervals, saccade velocity, and other thresholds (Duchowski,
2007). Although custom implementations are always possible, many users will adopt
saccade and fixation detection approaches, along with corresponding thresholds, di-
rectly from software tools provided with eye-tracking hardware.
Identifying eye-movement features is only the first step in an eye-tracking study.
As where the user's eyes are looking and what they are looking at on the screen are
both important (Jacob and Karn, 2003), appropriate use of eye-tracking data often
requires mapping eye-gaze data to screen coordinates (Duchowski, 2007), and then
integrating that data with information regarding the contents of the screen display at
each time point and any additional interaction about mouse and keyboard interaction.
Software tools that automatically synchronize these data streams can simplify the
data interpretation process (Crowe and Narayanan, 2000). Systems that can overlay
“trails” indicating the path of a user's gaze onto screen shots can be particularly use-
ful (Figure 13.1). As data analysis tools are often tied to specific hardware platforms,
eye-gaze research studies should be carefully designed and controlled (Duchowski,
2007), so as to minimize the risk of artifacts in data collection and interpretation that
might influence interpretation and results.

13.2.2 APPLICATIONS
When interpretation and analysis challenges are handled appropriately, eye-gaze data
can present researchers with intriguing possibilities. If we can understand how users
move their eyes when completing various interface tasks, we might gain some insight
into where attention is focused and how choices are made. This additional data can
take us beyond the relatively uninformative traces of mouse and keyboard events, fill-
ing in the holes: just where did the user look before she moved her mouse from one
menu to the next? Which portions of a web page initially attract user attention?