Online Brain Computer Interface Based Five Classes EEG To Control Humanoid Robotic Hand

Abstract

The proposed system had three stages in general, first stage was feature extraction, second stage was training a machine learning algorithm and third stage was online feature extraction and classification of ME/MI to control HRH. Variation for two kinds of feature extraction methods were proposed, Autoregressive (AR) coefficients and Common Spatial Pattern (CSP). Principal Component analysis (PCA) was used to reduce the dimensionality of AR feature. The output of the two methods were concatenated and normalized to train Support Vector Machine (SVM) algorithm. During online stage, EEG signal was acquired using EMOTIV EPOC EEG headset and same processing steps were applied as in training phase. The trained SVM module was used to predict the class of motion from the acquired EEG signal with 97.5% of online accuracy with the aid of majority voting. The predicted class was used as online signal to move the HRH to its corresponding hand gesture.