Sub-band-power-based efficient Brain Computer Interface for wheelchair control

Abstract

An efficient Brain Computer Interface (BCI) is designed and implemented to allow disabled people to control the motion of wheelchairs. It uses a compact portable EEG sensor to capture 14 brain signals and wirelessly feed them to the PC. Four classes of motions are used: Forward, Backward, Left, and Right. The signals are obtained in a free-style manner without compelling users to perform pre-defined mental operations. This led to variations in the results that shed some light on the cognitive aspect of the problem. Principal Component Analysis (PCA) and Sub-Band Powers obtained from the Wavelet Transform are used to reduce the signal dimensionality from nearly 14000 to only 3. A Feed-Forward Neural Network with Back Propagation is used as a classifier. The average classification rate is 91 % on the overall and as high as 97.5 % for some users. The effect of mother wavelet type and user dependence are also investigated.