Inverse Eye Tracking – Inferring Eye Movements from Keypress Data

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Science | Master's thesis

Date

2024-12-27

Department

Major/Subject

Human-Computer Interaction

Mcode

Degree programme

Master's Programme in Computer, Communication and Information Sciences

Language

en

Pages

40

Series

Abstract

We propose a model that predicts where users direct their gaze during interactions using only keypress data. Starting from a log of keypresses, the model generates a scanpath that details the user’s eye movements over time as they type. This model serves as a substitute for human data in scenarios where collecting eye-tracking data is costly or impractical. Our approach is built on three key insights: first, we design an inference architecture that captures individual user characteristics as a low-dimensional parameter vector; second, we introduce a novel loss function to align inferred eye movements with the timing of keypresses; third, we adopt a hybrid training strategy combining human data with synthetically generated data. This method is applicable to interactive systems equipped with predictive models of user behavior. In our evaluation, focused on the demanding task of touchscreen typing, the model demonstrated high accuracy in replicating real eye movements.

Description

Supervisor

Oulasvirta, Antti

Keywords

gaze movement, touchscreen typing, amortized inference, eye-hand coordination, visual attention, computational modeling

Other note

Attachment notes Description: The videos for Individual-level analysis Attachments: supplementary_videos.zip

Citation