Visual target selection using brain signals acquired using Electroencephalography (EEG)

Abstract

Past studies had shown promising results of Electroencephalography(EEG) based BCI as tool for communication and control. Several channels of BCI such as motor imagery, steady state visual evoked potential and visual spatial attention had been proposed. In this project, the goal was to demonstrate EEG-based control to a robotic device (SPHERO) using a covert visual spatial attention task. Specifically, it used the alpha and beta band neural oscillations observed during the task as input features for training of model. Band power of those neural oscillation were used to train a Linear Discriminant Analysis model. Average validation accuracy of 72.5% and average test accuracy of 60.9% were obtained from five subjects. A decreased in accuracy was attributed to feature shift due to non-stationarity of features and increasing fatigue level. Future designs for BCI could explore ways to minimised fatigue induced during the course of experiment and adapt the classifier to account for feature shifts