Dataset 2: Monopolar EOG Data Recorded Under Stationary Head Pose Conditions

This work aims to propose a novel method to estimate the gaze from electrooculography (EOG) signals while compensating for the baseline drift, and which intrinsically detects fixations, saccades and blinks. In contrast to existing baseline drift mitigation techniques, the proposed framework is real-time-implementable, and does not require the average point of gaze to lie at the primary gaze position, nor does it disrupt the overall ocular pose-induced EOG signal DC characteristics. The EOG data used to validate the proposed method is also being made publicly available. Methods: The proposed method is based on the dual Kalman filter, which estimates the gaze angles (GAs) and the baseline concurrently, taking into consideration the EOG signal’s non-stationary and temporally-multimodal characteristics. In fact, it is a multiple-model technique based on a battery model of the eye wherein fixations, saccades and blinks are modelled separately. Results: When applied to short EOG data segments, a horizontal and vertical GA estimation error of 1.64 ± 0.82°and 1.97 ± 0.34°, respectively, was obtained, which compared well with the corresponding results obtained using the state-of-the-art linear regression models. Conversely, for longer data segments, the proposed framework yielded superior GA estimation performance when compared to the state-of-the-art techniques. Eye movement detection and labelling F-scores exceeding 90% were achieved. Conclusion: The proposed method yields reliable gaze estimation performance, and accurately detects fixations, saccades and blinks.