Design and development of a motor imagery based interfaces of wheelchair in a simulated virtual environment

Abstract

Patients suffering from diseases like motor neuron diseases (MND), or trauma such as spinal cord injury (SCI), and amputation are not able to move. Presented is a work on combining the power wheelchair designed to aid the movement of disabled patient and a Brain Computer Interface (BCI) can be used to replace conventional joystick so that it can be controlled without using hands. By using the BCI, the brain signal emanated during Motor Imagery (MI) tasks can be converted into control signal for power wheelchair maneuvering. In this research, five subjects are requested to perform six Kinesthetic Motor Imagery tasks plus one relax task and the Electroencephalography (EEG) signals are recorded. Elliptic filter was used to remove power line noise. The proposed feature, combined feature of Fractal Dimension with Mel-frequency Cepstral Coefficients has outperformed the others. It was able to improve the classification performance to a satisfactory level especially for the subject 3 which yielded relatively poor result by using four other feature extraction methods. The classifiers network parameters were experimentally selected and the Levenberg-Marquardt training algorithm was used to train the networks. The Multilayer Perceptron Neural Network (MLPNN) outperformed Elman Recurrent Neural Network and Nonlinear Autoregressive Exogenous model (NARX) with average accuracy of 91.7%. The developed network models was further tested and evaluated with two simulated virtual environment created by using MATLAB graphical user interface (GUI). The simulation results suggested that step by step control is better than continuous control of wheelchair, and also the proposed feature, combined feature of FD with MFCCs and MLPNN can be used to classify Motor Imagery signal for directional control of powered wheelchair.