A systematic quantitative analysis on bipolar channel selection for EOG-based gaze displacement estimation

Abstract

The conventional electrode configuration that is generally used for electrooculography (EOG)-based gaze estimation comprises two bipolar channels, which are computed from the EOG signals recorded from two horizontally-aligned electrodes next to the outer canthi and two vertically-aligned electrodes attached superiorly and inferiorly to one of the eyes, respectively. While other unconventional electrode configurations have also been sometimes used in the literature, there are limited works on how these compare to the conventional configuration. Hence, the main contribution of this work is to analyse the flexibility in the electrode placement for EOG-based gaze estimation. Specifically, the gaze displacement estimation (GDE) performance obtained using different single-bipolar-channel and two-bipolar-channel configurations was analysed, and, in the case of the latter, the GDE performance was statistically compared to that obtained using the conventional configuration. In contrast to other comparative works reported in the literature, this work carries out an exhaustive analysis by systematically analysing all the different configurations that may be computed from a setup with 16 electrodes attached at equiangular positions around the eyes. Furthermore, the eye movements were carried out systematically in isotropic directions, and the EOG dataset used in this work is also being made publicly available. The results obtained demonstrate that the conventional horizontal and vertical bipolar EOG channels are not the only configuration that could be used for EOG-based GDE, but there are various other configurations with two bipolar channels that yield a performance that is statistically equivalent or even better than that of the conventional configuration.