Driving Behavior Study Based On Electroencephalography Data Analysis

In recent years,with the increasing number of car ownership in China,the number of car accidents and the resulting casualties and direct economic losses remain high.How to improve the road traffic safety level has become an important research topic in the field of traffic and transportation.Car driving is a complex activity that involves drivers' perception,judgment,decision-making,and operation.It requires the brain to coordinate drivers'driving operations.The electroencephalography(EEG)signal can reflect the psychophysiological state of the driver and thus characterize the driver's perception activity.Applying EEG data analysis to driving behavior research can help explain the driving behavior generation mechanism from the new perspective of cognitive neuroscience,and bring new solutions to traffic safety problems.Driving behavior research is the core research content in the field of road traffic safety.The identification and prediction of driving behavior status is of great significance to the development of intelligent assisted driving system and the improvement of road traffic safety level.Through simulated driving experiments,this paper synchronously collected drivers' EEG data and driving behavior data,and used the knowledge of cognitive neuroscience and machine learning to analyze micro-driving behavior characteristics,explore the relationship between drivers' EEG activity and driving behavior and further build driving behavior recognition and prediction models.The main research contents are as follows:(1)Micro-driving behavior analysis and influencing factorsThe driving simulation scenarios of the daily car-following,lane change and overtaking were established.The driving behavior data was collected through the simulated driving experiments and the key variables in drivers' car-following,lane change and overtaking process were extracted.Taking traffic state,traffic density and direction as the influencing factors,the effects of different factors on lane change,overtaking and car-following behavior were studied by means of repeated measures analysis of variance and Friedman test.The results indicated that with the increase of traffic density,the drivers' lane change and overtaking frequency increase as well as the lane change and overtaking intention increase;and the initial space headway and time headway of overtaking decrease with the increase of traffic density,thus affecting overtaking security.As for the direction selection of lane change and overtaking,the drivers have no significant preference.However,the research data showed that the right overtaking has a shorter overtaking duration,a smaller initial head spacing and headway distance,and a larger initial acceleration,and the field of view is restricted,resulting in high driving risks.(2)Exploring the relationship between EEG and driving behaviorBased on the EEG and driving behavior data collected by the simulated driving experiment under ordinary driving scenario,the power spectrum analysis was used to extract the amplitude,log-transformed power and power spectral density of each brain wave in each brain region as the EEG features;the original driving behavior data was used to extract acceleration,space headway,speed,time headway,lane deviation,and amplitude of steering wheel movements as the driving behavior features.The correlation analysis index system established by Pearson correlation coefficient was used to explore the correlation between EEG and driving behavior from four aspects:brain regions,brain waves,EEG features and driving behavior features.The results indicated that the driving activity is a complex behavior that requires the coordinated participation of the four major brain regions,especially the temporal,occipital,and frontal regions,? wave,as well as log-transformed power of ? wave,was found to be most relevant to the ordinary driving behavior.Furthermore,acceleration,speed,and space headway may have potential correlation with EEG features.(3)Driving behavior recognition based on EEG dataA simulated car-following driving experiment was designed and conducted to simultaneously collect drivers'driving behavior data and EEG data.Next,a two-layer learning method for driving behavior recognition using EEG data was established.In Layer I,two-dimensional driving behavior features representing driving style and stability were selected and extracted from raw driving behavior data using K-means and support vector machine recursive feature elimination.Five groups of driving behaviors were classified,including aggressive-stable,unaggressive-stable,unaggressive-unstable,aggressive-unstable and normal groups.In Layer II,the classification results from Layer I were utilized as inputs to generate a K-Nearest-Neighbor classifier identifying driving behavior groups using EEG data.Using independent component analysis,a fast Fourier transformation,and linear discriminant analysis sequentially,the raw EEG signals were processed to extract two core EEG features.Classifier performance was enhanced using the adaptive synthetic sampling approach.A leave-one-subject-out cross validation was conducted.The results showed that the average classification accuracy for all tested traffic states was 69.5%and the highest accuracy reached 83.5%.(4)Driving state prediction based on EEG feature extractionThe EEG-based driving behavior recognition model was adjusted to the short-term driving state prediction model,and various EEG analysis methods(independent component analysis and brain region source localization)and signal processing methods(fast Fourier transform and wavelet packet transform)were investigated.The prediction performance of driving features,EEG features and hybrid features of them was evaluated and compared.The results showed that the transformed forms of EEG,including log-transformed power,power spectral density,relative energy and Shannon's entropy,have better performance than the amplitude and energy of EEG.Next,the full brain region features were found to have better prediction performance than any single brain region features and also outperform the independent component features.Finally,in the comparison of all input features,EEG-based model has better performance than driving-data-based model(i.e.,83.84%versus 71.59%)and the integrated model of driving features and the full brain regions features extracted by wavelet analysis outperforms other types of features with the highest accuracy of 86.27%.