Research On Driver Status Online Recognition Simulation System Based On Virtual Reality

On-line recognition of drivers’ alertness is of great significance to reduce road traffic accidents and ensure the safety of people’s lives and property.In this dissertation,a virtual reality simulation driving experiment is built to synchronously collect the driver’s EEG data and driving behavior data.Using cognitive neuroscience,machine learning and other related knowledge,this dissertation studies the correlation between driver’s behavior and EEG state.The classification,recognition and real-time monitoring of driver status are deeply studied based on EEG signals.The main research contents of this dissertation are as follows:(1)Experimental design and data acquisition.Two experimental tasks were designed: driving in a straight line and braking on the sidewalk.A simulated real virtual reality driving scene is built through Unity 3D and HTC Vive.At the same time,Neuroscan64 guided EEG acquisition equipment is used to collect EEG data and driving behavior data during driving,so as to analyze the driver’s alertness state in the later stage.(2)Correlation analysis between EEG data and driving behavior.The characteristics of collected EEG signals and driving behavior data are extracted,and the Pearson correlation coefficient is used to calculate the correlation between them.It is found that the vehicle offset and steering wheel angle have the strongest correlation with EEG activity,and the temporal region and rhythms band have the strongest correlation with driving behavior data.(3)Classification of driver state alertness.By analyzing the vehicle state data most related to EEG signals,K-means clustering algorithm is used to classify the three corresponding states of drivers.T-test is used to detect the mean and standard deviation of vehicle offset and steering wheel angle in the classified vehicle state data.It is found that the vehicle state data classification using K-means clustering satisfies the T-test.Then,features of the classified EEG data are extracted,including 8 channels in the temporal region,7 sub-bands and 4 features after the wavelet transform.Finally,the 3-class classifications accuracy under different feature vectors is obtained by using SVM and RF.It is shown that the feature vector combined with four channels(P8,T7,T8,P7)under(9 sub-band CV feature has higher classification accuracy,and the accuracy of SVM classifier is higher than that of RF classifier.(4)On-line recognition of driver status.On-line recognition is carried out in this dissertation,Neuracle portable EEG acquisition equipment is used to collect EEG signals of four channels(P8,T7,T8 and P7).CV feature vector under(9 sub-band and SVM classifier are used to realize 3-class on-line recognition of drivers’ status,i.e.,high alertness,half alertness and low alertness.It can be seen that the average accuracy of online recognition is 86.7655%.