| For the driver,the process of driving a vehicle requires the perception of the surrounding environment,driving decisions and vehicle manipulation,so the driver will withstand a certain amount of mental load.When the driver’s mental load is too low or too high,it is easy to cause decision-making or operational errors,resulting in traffic accidents.Therefore,the detection of driver mental load state is of great significance to the development of driver mental load monitoring system and the improvement of road traffic safety.Most of the previous driver mental load evaluation models are mental load state classification,and the mental load state is defined by the driving scene classification label,which can not truly reflect the driver’s mental load.Therefore,this paper comprehensively considers the subjective mental load feelings of drivers,and uses physiological characteristics,vehicle handling characteristics and other characteristics as inputs to identify the mental load state of drivers.Firstly,driving simulation experiments containing four driving scenarios were designed and carried out.Through the experiments,vehicle control data,EEG and ECG signals of drivers in each driving scenario were collected,and subjective load scores of drivers in each scenario were collected by NSAS-TLX table.Statistical methods were used to analyze the differences of physiological characteristics and vehicle handling characteristics in different scenes.The results showed that most of the characteristics had significant differences among different scenes.Compared with normal driving in straight road,driving on curves or driving with secondary tasks can increase the driver’s mental load,and the influence of driving with secondary task has greater effect on the driver’s mental load.Secondly,KNN,RF and XGBOOST algorithms were used to construct the driver mental load classification and evaluation model,and the classification effect of the model under different feature combination input was compared and analyzed.Then,the driving performance of drivers in four driving scenarios is deeply analyzed.Combined with the model classification results,it is illustrated that the traditional way of defining mental load classification labels based on driving scenarios has limitations,and the problem of wrong definition of classification labels may occur,which affects the model recognition effect.Finally,two improved mental load assessment methods are proposed to solve the problem of the wrong definition of classification labels.One is based on the driver’s subjective load evaluation and driving performance correction error classification label,using the revised data set to carry out the training of classification model.Another approach is to consider the driver’s subjective mental load feelings,build regression evaluation model based on machine learning algorithmsl,physiological signal characteristics and vehicle handling characteristics as input variables,the driver’s NASA-TLX subjective load rating as output variables,to evaluate the driver’s mental load.The results show that the two improved mental load evaluation methods proposed in this paper can effectively identify the driver’s mental load,and can eliminate the difference of the impact of the tasks on the driver,and avoid the problem of labeling error.The research content of this paper can provide a basis for the development and design of driver condition monitoring system and driver assistance system,and help to improve the level of road traffic safety in the future. |
