Classification of Electroencephalogram Signals for Human Motor Actions

Abstract

In recent years research into electroencephalograph (EEG) based Brain Computer Interfaces (BCI) have focused on imagined hand and body movement. In contrast, current studies of actual hand movement tend to simply apply techniques for imagined movement directly onto actual movement without adjusting for the possibility of difference in EEG signals between actual and imagined action. This study aims to find a set of parameters, algorithms and acquisition techniques to maximize the classification accuracy for mapping EEG signals onto actual hand actions. Data is directly collected from subjects measuring their hand actions (using a set of VR gloves) and neuro-physical signals (using EEG and EMG sensors) for four different hand actions. This data is then preprocessed and features selected using the method of Common Spatial Patterns (CSP). These features are then processed using a number of classification algorithms. Accuracies up to 74.2% have been achieved, showing that there is an optimal set of parameters.