Recognition Method Of Drinking Driving Behavior Considering Driver's Propensity

With the rapid development of Chinese vehicle manufacture,the number of vehicles ownership has increased year by year,and vehicles have become the obbligato travel tools.Although it's convenient for us to travel by vehicles,vehicles also cause some serious traffic safety problems at the same time.Drivers are important parts of the road traffic system.In road traffic accidents,traffic accidents caused by the drivers' dangerous driving behaviors occupy large proportion.And drinking driving is a representative dangerous driving behaviors which is seriously harmful to traffic safety.So it is of great significance to prevent the occurrence of drunk driving by exploring the accurate,automatic,intelligent and active recognition method of drinking driving.The characteristics of driving behaviors data,such as the root,real time and low intrusiveness,make the active recognition method of drinking driving based on driving behaviors become a hotspot in traffic safety research.The differences of driving behaviors caused by different physical and psychological characteristics of drivers are important reasons for reducing the accuracy of traditional recognition method of drinking driving based on driving behaviors.Drivers' driving propensity can comprehensively reflect differences of driving behaviors caused by gender,age,driving age and driving experience.So it's of significance for improving the identified accuracy of drinking driving to consider driving propensity during the exploration of recognition method of drinking driving based on driving behaviors.Firstly,the driving propensity of drivers were determined by questionnaire tests,and all drivers were divided into three types: radical,common and conservative.Different driving propensities drivers' driving behaviors data before and after drinking were obtained by the simulated driving experiments,and drinking driving behaviors' characteristics of drivers with different driving propensities were studied by statistical methods.Secondly,the initial driving behaviors parameters' demensionalities were reduced by factor analysis,for the general drivers whose driving propensity were not be considered and the common factors were extracted as the input vector of multilayer neural network.The neural network was trained by error back propagation algorithm to establish recognition model of drinking driving based on factor analysis and multilayer neural network.Thirdly,the neural network classifier was trained by different driving propensities drivers' driving behaviors data under different drinking levels so that the classification accuracy of alternative parameter sets were obtained.And the classification accuracy of alternative parameter sets were inputted into discrete particle swarm optimization to extract drinking driving's characteristic parameters of drivers with different driving propensities.Lastly,the influence of driving propensity on the recognition model of drinking driving was importantly analyzed.The drinking driving characteristic data of drivers with different driving propensities was used to establish recognition model of drinking driving based on dynamic bayesian network which considers driving propensities.The recognition result of model was verified by experimental data,the result showed that the recognition model of drinking driving considering driving propensities can accurately recognize drinking driving behaviors.In comparison with the recognition model of drinking driving which doesn't consider driving propensities,the accuracy of this model which consider driving propensities is improved.This paper can enrich the study of dangerous driving behaviors and provide some references for the development of more advanced active recognition methods of drinking driving.