Thesis Defense

 

"Democratizing Eye Tracking

Alexandra Papoutsaki

Thursday, July 27, 2017, at 1:00 P.M.

Lubrano Conference Room (CIT Room 477 - 4th Floor)

Eye tracking, the process of capturing the location of the gaze within a display, is extensively used in usability studies, psychology experiments, human-computer interaction research, and a broad range of other applications. The setup and operation of modern eye trackers are time-consuming and a specialist is needed to calibrate them and be present throughout the experiment, leading to highly-controlled user studies with artificial tasks and only a small number of participants. In addition, their steep price, which rises to tens of thousands of dollars, restricts their use to only a small number of labs that can afford them.

This thesis presents a novel approach that aims to democratize eye tracking by using common webcams already present in laptops and desktops. We introduce WebGazer, a webcam eye tracker that infers the gaze of web visitors in real time. WebGazer is developed as an open-source JavaScript library that can be incorporated into any website. Its eye tracking model self-calibrates by mapping eye features to positions on the display that correspond to user interactions.

We investigate whether webcam eye tracking can lead to similar conclusions with in-lab eye tracking studies. We explore this question in the context of web search, by extending WebGazer so that it can predict the examined search element within a search engine result page. We use SearchGazer to replicate three seminal studies in the area of information retrieval and demonstrate that scalable and remote eye tracking studies on user behavior are possible at a fraction of cost and time.

Finally, we create a benchmark for webcam eye tracking with data collected from a lab study with more than 60 participants. This dataset allows us to investigate the relationship between user interactions and gaze, confirming past findings on the alignment of gaze with clicks and cursor movement, and introducing novel insights into the differences in gaze behavior across users based on their ability to touch type. Taking advantage of the temporal alignment of gaze and user interactions, we perform improvements in WebGazer's accuracy and functionality.

These contributions make eye tracking accessible to everyday users, researchers, and developers. Traditional eye tracking studies that are confined in labs can now be performed remotely and at a scale. Subjects can participate in studies in their everyday environments which can yield a more naturalistic behavior and lead to more powerful insights.

Host: Jeff Huang