00000ctmaa22000004a 4500 UP-99796217608682692 Buklod 20090203093843.0 m |o d | ta 090203s2008 xx d r |||| u| (iLib)UPD-00073797419 DENGII eng DMLUC LG 995 2008 E64 B38 Bautista, Mario Miguel A. Effects of a dynamic reference frame in offline classification of imagined movements for EEG-based brain-computer interface Mario Miguel A. Bautista. 2008. [vii], 127 leaves ill. "October 2008." Thesis (M.S. Electrical Engineering)--University of the Philippines Diliman. The objective of modern day Brain Computer Interface (BCI) research is to achieve high accuracy in classifying the inputs from mental activities. However, majority of past studies have concentrated on classification using a controlled environment, i.e. using static reference frame (SRF). With traditional researches using SRF, users are asked to gaze at a fixed point where no moving entities can distract the user. A previous research by Abundo and Sison investigated the effects of Dynamic Reference Frame (DRF) on mental task classification for EEG Based BCI. To simulate the DRF a video screen is used to inject the subject with camera rotation and zooming to simulate movement. In that study it was concluded that the imagined movement task achieved a two-way classification accuracy of 100% for SRF and 90% for DRF. The main objective of this research is to further validate the classification accuracy of different imagined movements (left arm raising, right arm raising, left leg raising and right leg raising) and its effect on the accuracy of the BMI system with the SRF to DRF switch. To achieve the objective the following methods and parameters were used: (a) 5 users performed 5 different Mental Tasks concentrating on imagined movement activities, (b) EEG provided a means to measure the scalp electrical activity, (c Power Spectral Density was used to reduce the EEG readings into feature matrices, (d) Feed forward neural network with back propagation was used for classification of the feature matrices. The results verified the validity of using Imagined Movement Activities for two-way classification with accuracies in the range of 90% to 100% for both SRF and DRF. The results also indicated that the SRF and DRF produced only a 3% to 5% difference in the accuracy. The accuracy of the classifier for five-way classification varied for different subjects. Varying the nodes indicated that the highest accuracies were produced at nodes equal to 6. Varying the training sizes indicated that the highest accuracies were reached at training size equal to 40, which is equivalent to 20 seconds worth of mental task training. Two test subjects posted accuracies of greater than 90%, which is significantly acceptable considering that the training for each subject consisted of only one test run. Brain-Computer interfaces Electroencephalography. Static Reference Frame (SRF). Dynamic Reference Frame (DRF). UP FI Thesis UPD DARCHIVES LG 995 2008 E64 B38 UPD DENG-II LG 995 2008 E64 B38 Thesis