Finite state machine and eog artifacts in eeg signals for wheelchair navigation / Roziana Ramli

• Main Author: Roziana, Ramli
• Format: Thesis
• Published: 2014
• Subjects: RE Ophthalmology; T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/7804/2/Thesis_Correction_(Roziana).pdf; http://studentsrepo.um.edu.my/7804/

Supporting physically disabled society with severe motor disabilities is very challenging as their needs differ depending on the severity of the impairment incurred. As an attempt to support them, a BCI for wheelchair navigation is developed to help them regain some mobility. In this study, a hybrid BCI that combine inputs from EEG and EOG signals for a more effective interface is proposed. Specifically, one EEG signal at O2 and two EOG artifacts embedded in EEG signals at C3 and C4 are used as inputs to an asynchronous wheelchair navigation system. Cz is taken as reference and the signals are all recorded using g.mobilab amplifier from 20 participants. The alpha rhythm extracted from O2 signal is related to eyelid position that determines whether the eyes are closed or open, while the delta rhythms extracted from C3 and C4 signals are related to horizontal eyeball movement used to infer the gaze direction. A sliding window is utilized to position important cues in the EEG signals at the center of the window to extract consistent features for accurate classification. The features from the O2 signal are variance, 2nd order difference plot and area. They are classified by thresholding and CTM. The delta rhythm data can be used directly as inputs to an LDA or K-Means classifier. Otherwise, a feature like area can be extracted from the delta signal and classified by thresholding. The system is modeled as a finite state machine with two modes, each containing three states. The transition between states is determined by fuzzy logic. This is to allow the wheelchair to move FORWARD and BACKWARD in six different directions with minimum error. In Online Session, the performance of various features and classifiers used are recorded and compared. A combination of features and classifier that achieve the highest accuracy is then implemented in the Navigation Session which are variance (98%) in alpha rhythm and K-Means (98%) in delta rhythms. Then, performance of the system is tested in Navigation Session. Only five participants who obtained the score higher than 98% in the Online Session were invited to perform the actual navigation tasks by maneuvering the wheelchair along two designated routes. High average performance rates of 98% for Route 1 and 96% for Route 2 were recorded and the participants managed to complete the tasks without collisions. This experiment also tested the usability of the BACKWARD movement when the wheelchair was trapped at tight dead ends with no space to make u-turn. The main contribution of this research work is in the right selection of the EOG artifacts and EEG signals used, the choice of the model that allows FORWARD and BACKWARD wheelchair movement and the fast execution time. Finally, the implementation of sliding window helps increase the performance rate.