Brain-Computer Interfaces (BCI) can be utilized to control a variety of devices, such as screen cursors, wheelchairs and robots. Noninvasive EEG-based BCI is an effective, albeit somewhat non-intuitive method. Gaze-based interfaces provide more natural and intuitive interfaces. However, they suffer from the Midas Touch problem, where targets and functions are unintentionally selected because a gaze-only based system cannot distinguish between whether a glance signifies intent or is just associated with looking around. In this thesis, we first describe a pure EEG-based BCI system that utilize common spatial patter filtering for feature extraction and linear discriminative analysis for classification. We achieved online feedback control over a 1D cursor control task on six subjects with a...[ Read more ]

Brain-Computer Interfaces (BCI) can be utilized to control a variety of devices, such as screen cursors, wheelchairs and robots. Noninvasive EEG-based BCI is an effective, albeit somewhat non-intuitive method. Gaze-based interfaces provide more natural and intuitive interfaces. However, they suffer from the Midas Touch problem, where targets and functions are unintentionally selected because a gaze-only based system cannot distinguish between whether a glance signifies intent or is just associated with looking around. In this thesis, we first describe a pure EEG-based BCI system that utilize common spatial patter filtering for feature extraction and linear discriminative analysis for classification. We achieved online feedback control over a 1D cursor control task on six subjects with average hit rate of 85.6%. We further detailed a multi-modal HMI system integrates EEG and gaze information for constructing and testing hybrid human machine interfaces. Using this system, we show that integrating EEG information can enhance the controllability of a gaze-based cursor control system. The combined EEG/gaze based control gives significantly finer control over the cursor trajectory, improving the directness of target reaching by 17% and reducing unintentional collisions by 36%.

Keywords – Brain Computer Interface, Human Machine Interface, EEG Signal Processing, EEG Classification