Brain Computer Interface For Virtual Navigation Based On Motor Imagery And Alpha Waves

Brain-computer interfaces (BCIs) is an artificial communication and control system, which offers an alternative pathway between the brain and the external machine, instead of the normal peripheral nerves and muscles. By signal utilized in BCIs, BCIs can be divided into invasive and non-invasive BCIs. BCIs based on motor imagery (MI) are a kind of non-invasive BCIs. For BCIs based on motor imagery, classification of 4-class imaginary tasks has been realized in off-line research. But for the online system, most of the research results stayed in classification of 2-class motor imagery and the dimension of controlling is thus limited. In order to improve the performance of the multi-direction online control of MI-based BCI, there are still lots of difficulties:limited distinguished directions, poor online performance and high dependence on subjects. So, this study proposes two online BCI designs to realize multi-direction controlling, improve online performance and lower dependence on subjects.In the first system, EEGs related to four kinds of tasks-imagine left hand, right hand, feet movement and rest-are decoded to control rotating towards three directions and not in control(NC) state. However in the off-line analysis and on-line experiments, the classification accuracy is 25%, which is not good enough for a stable system. In the second system, alpha waves produced by closing eyes is used as state transition switch. A threshold is set in the online system. When rising alpha wave exceeds the threshold, which means subjects want to change state, subject can choose the state he/she wants to control, which includes left-right movement and forward-backward movement. Then subject can control movement using motor imagery of left or right hand. The biggest advantage of the system is to use EEGs of 2 motor tasks to control motion in 4 directions. In the online experiments of the second system, three of five subjects without motor imagery experiment experience perform quite well in the first experiment with an average accuracy of 77.64%(the best is 88%). Besides, the best average classification accuracy of the five subjects is 93.25%. That means our system reduces the dependence on subjects. Based on the achievement of previous studies, I established a hybrid BCI for virtual navigation based on motor imagery and close-eyes alpha, which exceeds the performance of using motor imagery alone.