Research On Virtual Human Control Method Based On Brain-computer Interface

Brain-Computer Interface(BCI)can express brain activity information by controlling external devices such as medical equipment,wheelchairs,and manipulators,so as to achieve the purpose of restoring human capabilities.Even so,most BCI technology is still in the experimental stage,and algorithm isn't perfect.At the same time,due to the high cost of external equipment,complex connections,and high consumption of artificial intelligence algorithm,the recognition accuracy of EEG signals cannot be guaranteed.Therefore,it is of great significance to use virtual human for algorithm research.For this reason,the technology of motor imaging EEG signal recognition and control virtual human is investigated in this study.The channel selection,feature extraction,classification and recognition of EEG signal are mainly studied.The following are the main conclusions of the study:(1)Due to the large number of EEG signal channels,the similarity and the low recognition efficiency,the method combining mutual information and random forest is used to select the EEG signal channels.First,mutual information is used to remove similar channels.Then a random forest is used to select the most important channels.This method can take into account the various relationships between channels,obtain the optimal channels required for high precision.It improves the subsequent recognition effect.(2)For EEG signals with small differences,weak amplitudes,and strong inter-channel correlations,the state space model is used as a feature model of the EEG signal.The parameters of the state space model are extracted as the features of the EEG signal.The EM algorithm is used to identify the parameters of the state space model.The state space model can not only extract the spatiotemporal features of EEG but also establish the nonlinearity relationship between channels.The method improves the effect of subsequent identification.(3)In view of the timing and repeatability of EEG signals,important timing and features need to be weighted to improve the accuracy.The SE-GRUTA model is proposed to identify EEG signals.Before the feature is input into the GRU model,the squeeze excitation operation is performed.Then the output weight is calculated each time step in the GRU model.Finally,the final classification result is obtained through the full connection.(4)A virtual human control system based on Web GL is Developed,the virtual human is controlled by EEG signal.Maya is used to create a virtual human and simulation environment.The virtual human is rendered in the system page through Web GL.The model is applied to this system to realize the recognition of brain wave signals.Corresponding actions are realized by the code of virtual human motion.Thereby,the EEG signal controls the virtual human to perform movement.