The Viability of Using Remote Webcam-Based Eye Tracking to Monitor Attention Allocation in Educational Research

• In Event: Remote Measures of Attention and Affect: Online Data Collection in the Age of COVID-19

Abstract

Eye tracking has long been used as a means of monitoring the allocation of overt visual attention in lab-based cognitive and user-experience research (Rayner, 1998). It has also played an important role in educational research and the learning sciences as it provides insight into the learning process (Dahlstrom-Hakki et al., 2019). Eye tracking research primarily relies on desktop or head-mounted infrared cameras that are able to monitor pupil movement and/or corneal reflections and are generally conducted in a lab setting where head movements are minimized and lighting conditions carefully controlled. In recent years however, there has been a proliferation of webcam-based eye tracking solutions. Here we review the most promising of these solutions and their viability as a means of assessing attention allocation in educational research.

The viability of any eye tracking solution for research purposes is generally based on a device’s spatial and temporal accuracy. Spatial accuracy refers to how closely the eye tracker’s estimated gaze location is to the actual gaze location and temporal accuracy refers to how closely the timing of a recorded eye movement is to the actual event. Webcam-based solutions have generally faired poorly on both these metrics but have seen significant improvement as hardware, and calibration and estimation algorithms have improved. Recent webcam-solutions have relied on one of two approaches, software that uses trained estimation algorithms that a) map pupil images directly to screen coordinates (Papoutsaki et al., 2016); or b) map pupil images to a model of the eyeball as a means of estimating a gaze vector (Baltrusaitis et al., 2018). This work directly compares the spatial and temporal accuracy of both of these approaches endorsed by a selection of current web eye tracking tools and discusses their viability as an educational research tool.

A sample of participants with a range of ages was used to test spatial and temporal accuracy of these tools for use in data collection with both children and adults. Data collection included having participants move through an array of targets while their data was tracked using each of the webcam-based solutions. In addition, data was simultaneously collected from a research grade eye tracker for comparison purposes.

We will share findings from this work along with recommendations for the types of studies and research designs that are viable using the webcam-based tools. We will also discuss the technical challenges and limitations, ethical considerations, equity challenges, and data management and analysis needs of remote webcam-based eye tracking.

Authors

• Ibrahim Dahlstrom-Hakki, TERC

• Fengfeng Ke, Florida State University

• Ruohan Liu, University of Florida

• Zlatko Sokolikj, Florida State University