Research On New Technique For Mobile Device Control With Computer Vision Based BCI

Brain-Computer Interface (BCI) is a new information communication and control technique based on the direct channel between human brain and other external devices without resorting to routine passages like peripheral nerve and muscle. The Mobile device control, as an important application of BCI, has become one of BCI technical difficulties for its rigorous demand in terms of real-time and safety.At present, one-step operation dominates mobile device control through BCI, in which the frequent orders lead to difficulties of providing on-line command and monitoring external device's movement and local scene simultaneously. To cope with these difficulties, this thesis proposed and designed a visual evoked event-related potential P300-based on-line BCI system to control mobile devices (such as intelligent wheelchairs or robot nurses), which manages to achieve stable, reliable and quick control over mobile external devices by means of computer vision-based target positioning and robot autonomous navigation that allow operators to select target area for operation by simply staring at different areas of mobile device movement scene appearing on the evoke interface. Consequently, real-time and stable control over the mobile device is realized, and subject's operation has been substantially reduced.To accommodate subjects'individual differences, a scheme of automatic selection of individualized parameters was worked out, and a mobile external device BCI off-line drill and exercise platform was designed to match a variety of individualized parameters selection of subjects so as to support work of the on-line system. The individualized parameters mainly cover evoke patterns suitable for different subjects, optimal channel group, stacking fold of one-time operations, data processing time window and reduced sample rate, etc. Off-line experiments were done for operators to select individualized parameters and undergo necessary operation exercises, and off-line analysis results can be applied to the on-line system, which improves the system's adaptability to different individuals.Processing of EEG is the core of on-line BCI. A multi-feature classification algorithm was designed based on Relevance Vector Machine (RVM) for pattern recognition, combinatorial optimization of optimal electrode group based on Recursive Feature Elimination approach based on RVM, and Genetic Algorithm based on RVM, which effectively improves the real-time and adaptability of the system.On the foundation of algorithm study, this thesis has completed the design and realization of the mobile device on-line BCI control system based on computer vision, and the author launched on-line system test with 6 subjects after feedback exercises. The accuracy rate of subjects'real mobile device control experiment has reached above 84%, with 100% as the highest accuracy rate achieved. The experiment results proved that the system can greatly reduce subjects'operational input, and it offers convenient, direct and safe operation with friendly and comfortable human-computer interface which involves comparatively more human elements.The research achievement in this thesis provides a new idea and a new technique for realizing desired on-line mobile device BCI control. It is very promising to be applied in self-care and restoration of severely disabled people, and gain ground in further popularization in future.