Gam360: sensing gaze activities of multi-persons in 360 degrees

Eye gaze reveals one's attentional focus in social interaction. However, most existing gaze estimation systems can only sense gaze activities within a specific angular range for a particular individual, thus limiting their applicability in multi-person face-to-face scenarios. In this paper, we propose GAM360, a flexible system that enables 360-degree multi-person 3D gaze and screen point-of-gaze (PoG) estimation. Different from previous techniques, we build the system based on a bottom-up fisheye camera, which can sense gaze activities from users in different directions simultaneously. We develop a real-time multi-person 3D gaze estimation algorithm, along with a method for multi-person screen PoG estimation. Experimental results demonstrate that the system can accurately track the gaze of multiple users in 360 degrees, achieving a 3D gaze estimation angular error of 8.7 degrees +/- 4.9 degrees\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$8.7<^>\circ {\pm } 4.9<^>\circ$$\end{document} under free head movement. Furthermore, we validate the effectiveness of the system across various application scenarios, including group discussion, desktop environment, and classroom environment, providing new insights into ubiquitous gaze sensing.