Sliding window method for eye movement detection based on electrooculogram signal

Abstract

In the past few decades, biomedical signals have played important roles in assisting diagnosis for medical purposes. After the rose of brain-computer interfaces (BCI) and human-machine interaction (HMI) concept, biomedical signals such as electroencephalograph (EEG) and electrooculograph (EOG) begun to be implemented in control and communication systems. EOG, the signal resulted from eye movement, has been used to design various applications from drowsiness detection to virtual keyboard control. The key of the system developed from EOG signal is the detection system for every eye movement. In this study, a sliding window technique is proposed to make eye movement patterns easier be formulated and using overlap window to avoid local extrema when computing the feature. Evaluation of this method shows that combination of 0.5 s-window length and 25% overlap give 17% and 1% false discovery rate (FDR) in vertical and horizontal channel while the true positive rate (TPR) in both channel is 98% The combination of automatic-window and 25% overlap give a better accuracy with 99% and 100% TPR in the two direction while the FDRs are 22% and 1%.