Dissertation
Designing gaze-based interaction for pervasive public displays
Vorschaubild nicht verfügbar
Volltext URI
Dokumententyp
Text/Dissertation
Zusatzinformation
Datum
2018
Autor:innen
Zeitschriftentitel
ISSN der Zeitschrift
Bandtitel
Quelle
Verlag
Ludwig-Maximilians-Universität München, Fakultät für Mathematik, Informatik und Statistik
Zusammenfassung
The last decade witnessed an increasing adoption of public interactive displays.
Displays can now be seen in many public areas, such as shopping malls, and train stations. There is also a growing trend towards using large public displays especially in airports, urban areas, universities and libraries.
Meanwhile, advances in eye tracking and visual computing promise straightforward integration of eye tracking on these displays for both: 1) monitoring the user's visual behavior to evaluate different aspects of the display, such as measuring the visual attention of passersby, and for 2) interaction purposes, such as allowing users to provide input, retrieve content, or transfer data using their eye movements.
Gaze is particularly useful for pervasive public displays. In addition to being natural and intuitive, eye gaze can be detected from a distance, bringing interactivity to displays that are physically unreachable. Gaze reflects the user's intention and visual interests, and its subtle nature makes it well-suited for public interactions where social embarrassment and privacy concerns might hinder the experience.
On the downside, eye tracking technologies have traditionally been developed for desktop settings, where a user interacts from a stationary position and for a relatively long period of time. Interaction with public displays is fundamentally different and hence poses unique challenges when employing eye tracking. First, users of public displays are dynamic; users could approach the display from different directions, and interact from different positions or even while moving. This means that gaze-enabled displays should not expect users to be stationary at a specific position, but instead adapt to users' ever-changing position in front of the display. Second, users of public displays typically interact for short durations, often for a few seconds only. This means that contrary to desktop settings, public displays cannot afford requiring users to perform time-consuming calibration prior to interaction.
In this publications-based dissertation, we first report on a review of challenges of interactive public displays, and discuss the potential of gaze in addressing these challenges. We then showcase the implementation and in-depth evaluation of two applications where gaze is leveraged to address core problems in today's public displays. The first presents an eye-based solution, EyePACT, that tackles the parallax effect which is often experienced on today's touch-based public displays.
We found that EyePACT significantly improves accuracy even with varying degrees of parallax. The second is a novel multimodal system, GTmoPass, that combines gaze and touch input for secure user authentication on public displays. GTmoPass was found to be highly resilient to shoulder surfing, thermal attacks and smudge attacks, thereby offering a secure solution to an important problem on public displays.
The second part of the dissertation explores specific challenges of gaze-based interaction with public displays. First, we address the user positioning problem by means of active eye tracking. More specifically, we built a novel prototype, EyeScout, that dynamically moves the eye tracker based on the user's position without augmenting the user. This, in turn, allowed us to study and understand gaze-based interaction with public displays while walking, and when approaching the display from different positions. An evaluation revealed that EyeScout is well perceived by users, and improves the time needed to initiate gaze interaction by 62% compared to state-of-the-art. Second, we propose a system, Read2Calibrate, for calibrating eye trackers implicitly while users read text on displays. We found that although text-based calibration is less accurate than traditional methods, it integrates smoothly while reading and thereby more suitable for public displays. Finally, through our prototype system, EyeVote, we show how to allow users to select textual options on public displays via gaze without calibration. In a field deployment of EyeVote, we studied the trade-off between accuracy and selection speed when using calibration-free selection techniques. We found that users of public displays value faster interactions over accurate ones, and are willing to correct system errors in case of inaccuracies.
We conclude by discussing the implications of our findings on the design of gaze-based interaction for public displays, and how our work can be adapted for other domains apart from public displays, such as on handheld mobile dev