Reducing execution time for real-time motor imagery based BCI systems

Abstract

Brain Computer Interface (BCI) systems based on electroencephalography (EEG) has introduced a new communication method for people with severe motor disabilities. One of the main challenges of Motor Imagery (MI) is to develop a real-time BCI system. Using complex classification techniques to enhance the accuracy of the system may cause a remarkable delay of real-time systems. This paper aims to achieve high accuracy with low computational cost. Two public datasets (BCIC III IVa and BCIC IV IIa) were used in this study; to check the robustness of the proposed approach. Dimension reduction of input signal has been done by channel selection and extracting features using Root Mean Square (RMS). The extracted features have been examined with four different classifiers. Experimental results showed that using Least Squares classifier gives best results, compared to other classifiers, with minimum computational time.