Towards SSVEP-based, portable, responsive Brain-Computer Interface
[ 1 ] Instytut Automatyki i Inżynierii Informatycznej, Wydział Elektryczny, Politechnika Poznańska | [ 2 ] Wydział Elektryczny, Politechnika Poznańska | [ P ] pracownik
2015
referat
angielski
EN A Brain-Computer Interface in motion control application requires high system responsiveness and accuracy. SSVEP interface consisted of 2-8 stimuli and 2 channel EEG amplifier was presented in this paper. The observed stimulus is recognized based on a canonical correlation calculated in 1 second window, ensuring high interface responsiveness. A threshold classifier with hysteresis (T-H) was proposed for recognition purposes. Obtained results suggest that T-H classifier enables to significantly increase classifier performance (resulting in accuracy of 76%, while maintaining average false positive detection rate of stimulus different then observed one between 2-13%, depending on stimulus frequency). It was shown that the parameters of T-H classifier, maximizing true positive rate, can be estimated by gradient-based search since the single maximum was observed. Moreover the preliminary results, performed on a test group (N=4), suggest that for TH classifier exists a certain set of parameters for which the system accuracy is similar to accuracy obtained for user-trained classifier.
1095 - 1098
WoS (15)