Electrooculogram based blink detection to limit the risk of eye dystonia

Abstract

In this paper a system for detecting the possibility of eye dystonia, a neural disorder that causes a person to blink excessively, by eye movement analysis is proposed. The designed system counts the number of blinks for a particular time interval and thus detecting the risk of eye dystonia. Electrooculogram (EOG) signal is recorded to collect eye movement data using a laboratory developed acquisition system. Radial Basis Function(RBF) kernel Support Vector Machine (SVM) classifier and Feed forward neural network classifier is used to classify blinks from other types of eye movements using combinations of Wavelet coefficients, Autoregressive (AR) parameters and Hjorth parameters with Power Spectral Density (PSD) as signal features. A maximum average accuracy of 95.33% over all classes and participants is obtained using RBF-SVM classifier with a feature space of AR parameters of order 5 and PSD taken together.