Research On Distracted Driving And Information Filtering Strategy Based On Driving Performance

Traffic safety on road is a major global issue today.The main causes of road traffic accidents can be summarized into four basic elements: human,car,road,and environment.Driver behaviour is the leading factor affecting traffic accidents,among which distracted driving accounts for a considerable proportion.Changes of driving performance before and after the transition from non-distracted state to distracted state are analyzed using naturalistic driving experiment data,with considering factors like road type,traffic density,driving time,gender and age,to propose a model for driving distraction recognition.With the development of connected vehicle technology,various information provided by in-car devices can cause drivers to be distracted to varying degrees.Driving assistance strategies,such as "countermeasures against drivers' distraction",have emerged.Based on the theory of attention resource allocation,an information filtering strategy road test is designed,and a questionnaire survey is then conduceted to statistically analyze drivers' subjective feelings on information filtering strategy.Based on the driving simulator experiment of "information filtering" level design scheme,the effect of information filtering strategy on reducing driving distraction and improving traffic safety was evaluated.Firstly,by using sensor data from the naturalistic experiment,the time domain and frequency domain analyses are conducted to compare the changes of driving performance before and after the distraction task.The standard deviation of the deviation of lane-keeping,the standard deviation of the steering wheel angle and the standard deviation of the yaw rate significantly increased during distracted driving when compared to non-distracted driving.The average relative spectral power density distribution obtained by fast Fourier transform presents different laws,and the values of lane offset,steering angle,and yaw rate energy in frequency domain significantly increase after the start of non-driving task.No significant change was observed in the longitudinal control variables.Secondly,the driving performance variables with statistically significant changes are selected to constitute the structure of distracted driving indicators,so as to establish a support vector machine model for distraction identification combining the driving performance variables in time domain and frequency domain.Training sample and test sample sets are randomly seclected.And gaussian radial basis kernel function,parameters optimization algorithms like GSA,GA,and SMO,and a ten fold cross validation method are sequentially used.Eventually,the average accuracy of the SVM model of driving distraction recognition of GSA-SVM,GA-SVM and SMO-SVM are91.02%,93.14% and 90.67%,respectively.Additionally,likert scale is used to analyze the reliability and validity of the questionnaire contents.The principal component analysis is used to extract common factors and then the fuzzy comprehensive evaluation is conducted.Most of the drivers believe that the information filtering strategy can reduce the degree of driving distraction both in the peak period and off-peak period.The information filtering strategy has a more significant effect in the peak period than in the off-peak period,which is more helpful to reduce the degree of driving distraction and then improve driving safety.Lastly,the experimental data of driving simulator are used to analyze the effect of information filtering strategy on driving behavior when drivers were distracted by the interaction with in-vehicle applications.It is found that the traffic density can be used as the hypothesis variable of driving behavior in this experimental design.Only in the driving environment with high traffic density,the driver's attention resources are protected with information filtering strategy,when compared with those without information filtering strategy.Therefore,information filtering strategy can effectively reduce driving distraction,indicated by significantly reducction of lane deviation standard deviation(m)by about 23.8%(?=-0.056;95% confidence interval(0.235,0.291)),significantly reduction of the standard deviation of steering wheel angle(radian)by about 24.1%(?=-0.0079;95% confidence interval(0.0327,0.0406)),and significantly reduction of the standard deviation of acceleration(m/s~2)by about 16.2%(?=-0.0143;95% confidence interval(0.0882,0.1025)).