Recognition of EOG based reading task using AR features

Abstract

Eye movements play an important role in evaluating the process of reading. By visual inspection of the eye movements, it is possible to differentiate the reading process of different persons. The eye movements can be considered as objective tools for understanding the reading process. However, most of the eye movements are involuntary and out of our conscious control. Hence the reading process is better understood when the analysis of eye movements is automated. This research work presents a pilot study conducted in process of automating eye movement analysis to get an insight into the reading process. Electrooculogram (EOG) has been used for recording the eye movements from a group of 40 volunteers. Several autoregressive (AR) features based on Yule walker's method, Burg's method, modified covariance method and Linear Predictor Coefficients obtained using Levinson-Durbin recursion methods have been extracted from the raw EOG. The horizontal and vertical modes were then recognized by employing a recurrent Elman neural network. Simulation results show a classification accuracy of 99.95% which indicates the suitability of proposed scheme for human-computer interface applications.