Highly Accurate Two Channel Single-Cycle EOG Classification for Smart Wearable Technologies

Motion of the eye generates an electrical potential difference between the cornea and retina. This signal, known as the Electrooculogram (EOG), is unique and characteristic of eye movements. This can be leveraged for noninvasive eye-tracking. Although the accuracy of this procedure decreases as the number of ocular motions increases, it is crucial for enabling more complex tasks in the Human-Computer Interaction (HCI) field. The performance further deteriorates when the motion classification is constrained within a single EOG cycle. This is imperative and significant for real-time practical applications. In this experimental study, (i) we have addressed the complexities in working with real-time single cycle EOG signals and provided necessary solutions (ii) worked with nine distinct eye movements, and (iii) proposed a cascaded Artificial Neural Network (ANN) approach to classify nine different eye movements based on single-cycle EOG signal data. Leveraging statistical properties and dominant frequency analysis of the EOG signals, our approach achieved an accuracy of 99.26%, precision of 99.26%, recall of 99.32%, and a F1-score of 99.25%. This is a significant improvement over past studies conducted by various researchers for similar purposes and to the knowledge of the authors, such high-performance metrics over this wide range of eye movements based on single-cycle EOG signals have not been previously reported.