Research And Application Of Driving Intention Identification And Expression Method Based On Brain Computer Interface

Human factors are the main causes of traffic accidents,and Advanced Driving Assistance Systems(ADAS)can improve driving safety by predicting the driving behavior of drivers.All human behaviors are directed by the brain.In theory,EEG signals can reflect different driving intentions more directly and quickly.However,due to the difficulty of landing,most of the ADAS modules that have been practically applied at present indirectly understand the driver’s intention and behavior by analyzing some external physical signals or through image information.Studying the identification and expression methods of different driving intentions from the perspective of drivers’ EEG signals will help to better understand the physiological mechanism of drivers’ different driving intentions and behaviors,so as to help ADAS find the possible judgment or operation errors of drivers earlier and take countermeasures in advance,so as to increase the guarantee for the safety of vehicle driving.The main research contents of the paper are as follows:Firstly,a scheme for collecting and marking EEG signals of drivers under simulated driving conditions is designed,the main reasons for choosing the simulated driving experiment method are analyzed,and the software and hardware equipment for EEG acquisition,as well as the specific process and precautions of the experiment are introduced.Secondly,according to the source and characteristics of each artifact in the EEG signal,band-pass filter and Independent Component Analysis are used to remove the artifact components.By analyzing the EEG topographic maps of four driving intentions,such as left and right steering,acceleration and braking,it is found that the EEG rhythms of the four driving intentions are different in the spatial distribution characteristics,realizing the expression of different driving intentions through EEG signals,and enhances the interpretability of the driving intention recognition model in the paper to a certain extent.Then,based on this,the channel selection is completed,the feature of Filter Bank Common Spatial Pattern are extracted to reflect the above spatial feature differences,and the support vector machine method is used to classify the four driving intentions.By comparing the within-subject and cross-subject models,the feasibility and generality of the above algorithms are analyzed;By comparing the different selection methods of sampling time window and filtering frequency band,the individual differences of the formation time of driving intention and EEG rhythm are analyzed;Compared with other types of EEG features,the effectiveness of spatial features in expressing four different driving intentions is verified.Then,a convolutional neural network model with less complexity is built,and four different driving intentions are identified by using the end-to-end characteristics of deep learning.Gaussian noise is added to achieve EEG data augmentation,and the classification performance of the model is evaluated by confusion matrix and Kappa metric.Through comparative experiments,the significance of data processing strategies such as data augmentation,pre-filtering,convolution kernel size and pooling method,and activation function is analyzed.By comparing the classification performance of other types of deep neural network models,the advantages of this model in recognizing four different driving intentions are verified.Finally,a BCI system based on motor imagery is designed.By imagining three intentions,such as left and right turning actions and idle state,the controlled object is controlled to move in different ways online,which verifies the effectiveness of the FBCSP feature and the feasibility of online control of external objects through brain intentions.