12 May 2016 On the use of hidden Markov models for gaze pattern modeling
Author Affiliations +
Abstract
Some of the conventional metrics derived from gaze patterns (on computer screens) to study visual attention, engagement and fatigue are saccade counts, nearest neighbor index (NNI) and duration of dwells/fixations. Each of these metrics has drawbacks in modeling the behavior of gaze patterns; one such drawback comes from the fact that some portions on the screen are not as important as some other portions on the screen. This is addressed by computing the eye gaze metrics corresponding to important areas of interest (AOI) on the screen. There are some challenges in developing accurate AOI based metrics: firstly, the definition of AOI is always fuzzy; secondly, it is possible that the AOI may change adaptively over time. Hence, there is a need to introduce eye-gaze metrics that are aware of the AOI in the field of view; at the same time, the new metrics should be able to automatically select the AOI based on the nature of the gazes. In this paper, we propose a novel way of computing NNI based on continuous hidden Markov models (HMM) that model the gazes as 2D Gaussian observations (x-y coordinates of the gaze) with the mean at the center of the AOI and covariance that is related to the concentration of gazes. The proposed modeling allows us to accurately compute the NNI metric in the presence of multiple, undefined AOI on the screen in the presence of intermittent casual gazing that is modeled as random gazes on the screen.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Pujitha Mannaru, Balakumar Balasingam, Krishna Pattipati, Ciara Sibley, Joseph Coyne, "On the use of hidden Markov models for gaze pattern modeling", Proc. SPIE 9851, Next-Generation Analyst IV, 98510R (12 May 2016); doi: 10.1117/12.2224190; https://doi.org/10.1117/12.2224190
PROCEEDINGS
7 PAGES


SHARE
Back to Top