Multi-Eyes: A Framework for Multi-User Eye-Tracking using Webcameras

The human gaze provides informative cues on human behavior during interactions in multi-user environments. However, capturing this gaze information using traditional eye trackers often requires complex and costly experimental setups. Furthermore, conventional eye-tracking algorithms are catered for...

Full description

Saved in:
Bibliographic Details
Published inProceedings of the IEEE International Conference on Information Reuse and Integration (Online) pp. 308 - 313
Main Authors Mahanama, Bhanuka, Ashok, Vikas, Jayarathna, Sampath
Format Conference Proceeding
LanguageEnglish
Published IEEE 07.08.2024
Subjects
Deep Learning
Estimation
Eye Tracking
Gaze tracking
Hardware acceleration
Joint Attention
Optimization
Performance evaluation
Pipelines
Prototypes
Rough surfaces
Scalability
Throughput
Online AccessGet full text

Cover

More Information
Summary:The human gaze provides informative cues on human behavior during interactions in multi-user environments. However, capturing this gaze information using traditional eye trackers often requires complex and costly experimental setups. Furthermore, conventional eye-tracking algorithms are catered for single-user scenarios and cannot be used for multi-user environments. We propose Multi-Eyes, a commodity webcam-based solution offering scalability and cost-efficiency while leveraging the advancements in deep learning for capturing multi-user gaze. Multi-Eyes propose a three-step multi-user eye tracking framework that (1) detects gaze subjects, (2) estimates gaze, and (3) maps gaze-to-screen with a scalable, memory, and parameterefficient disentangled gaze estimation model. We evaluate the gaze estimation model using two publicly available datasets and the framework's utility through a joint-attention case study. Our proposed architecture achieves the lowest gaze error of 4.33, while the case study demonstrates the feasibility of the Multi-Eyes for multi-user interactions and joint attention with comparable results to the state-of-the-art.
ISSN:2835-5776
DOI:10.1109/IRI62200.2024.00069