Classification of Multi-class Motor Imagery for Control Application of BCI

Abstract

Electroencephalography (EEG) is an electrophysiological monitoring method to record the brain's spontaneous electrical activity over a period of time through multiple electrodes placed on the scalp. Other than its use in clinical procedures such as epilepsy detection, EEG is also used to drive Brain Computer Interfaces (BCIs). BCI is a communication system that uses only brain activity to control an external device without execution of any peripheral muscular activity. EEG based BCIs can use a variety of different electrophysiological signals/activities to determine the user’s intent. In this project, we employ a dynamic state called motor imagery to navigate a robotic assembly. Motor imagery (MI) is mental stimulation of a motor act such as movement of left hand or movement of right hand, without physically executing it. Such passive BCIs can be used to provide a tool for communication and navigation to paralyzed people. MI based BCIs require a necessary pre-processing technique, identification of the most distinguishing set of features and a suitable and efficient classifier to determine the user’s command from brain signals. Despite constant improvement in EEG acquisition, feature extraction and classification techniques for BCIs in the past two decades, the use of BCIs is very limited. Poor signal-to-noise ratio, temporal variations of brain signals, subject dependence and non-stationary nature of EEG signals hinder the performance of MI based BCI systems. This projects attempts to find an effective classification algorithm for accurate determination of user’s command which is then used to control the movement of a robotic assembly.