Date of Award

Summer 8-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Program/Concentration

Computer Science

Committee Director

Sampath Jayarathna

Committee Member

Michael L. Nelson

Committee Member

Michele C. Weigle

Committee Member

Vikas G. Ashok

Committee Member

Yusuke Yamani

Abstract

The human gaze provides informative cues for understanding behavior during multi-user interactions. However, eye-tracking in multi-user environments presents significant challenges, as conventional eye-tracking systems are typically limited to single-user setups and require a dedicated device for each participant. This results in costly, complex, and often restrictive experimental configurations due to the operational and hardware constraints of conventional eye trackers. In contrast, commodity hardware such as standard webcams offers a cost-efficient alternative for gaze estimation across multiple users in a co-located environment. Despite recent advances in computer vision, appearance-based gaze estimation, and the availability of large-scale gaze datasets, limited research exists on developing lightweight, cost-effective, commodity hardware-based eye-tracking systems capable of supporting multi-user environments.

To address the research gap, we propose Multi-Eyes, a real-time, multi-user eye-tracking framework designed for scalable and cost-efficient gaze estimation and analysis. Multi-Eyes integrates a three-stage pipeline encompassing user detection, appearance-based gaze estimation, and gazeto- surface mapping for gaze analytics. The system operates at 17 frames per second on standard hardware and achieves an average gaze accuracy of 319 mm (horizontal) and 219 mm (vertical) on a large-format display. Beyond raw gaze estimation, Multi-Eyes also enables the derivation of low-frequency fixation-based measures, demonstrating its potential utility in real-world interaction studies. At its core, the system features a parameter-efficient appearance-based gaze estimation model that achieves an in-domain gaze error of 4.95◦ using only 6.2 million parameters, 76% fewer parameters compared to ResNet-50 baselines while maintaining comparable performance. Additionally, the model incorporates an unsupervised domain adaptation strategy, enabling strong generalization and competitive performance across multiple unseen target domains. These contributions establish Multi-Eyes as a promising step toward scalable, low-cost multi-user eye-tracking systems.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/e27a-2j16

ISBN

9798293842360

ORCID

0000-0001-7773-7471

Download

Available for download on Sunday, September 19, 2027

Share

COinS