Driving Risk Identification Method Considering The Drivers' Physiological Characteristics Under Different Cognitive Workloads

Driving safety has always been a hot spot for research in the field of transportation.With the gradual application of Telematics and smart cars,the safety performance of vehicles is increasingly important.As it is difficult to realize fully automated driving in a short time,human-machine driving will be the main development direction for a long period of time in the future,and it is necessary to comprehensively grasp the physiological characteristics of drivers and driving status so as to effectively identify the risks of driving.Effective early warning and measures should be taken to avoid accidents and reduce casualties and property losses.It is helpful to dig into the physiological characteristics of drivers to identify the risks effectively,and improve the driving safety,and also provide theoretical basis for the development of driving assistant system.In this paper,we designed a real-world driving experiment and collected the driver's eye movements,EEG feature indicators and vehicle operation indicators.Cognitive load tasks of different difficulties were designed during the driving process,and drivers generated different levels of cognitive load in the process of calculating mathematical problems of different difficulties,thus realizing the comparison between the normal driving state and the driving state under the cognitive load task.The visual characteristics of drivers are investigated,and the driver's gaze and transfer characteristics under different cognitive loads are mainly studied.The visual gaze characteristics of drivers are macroscopically analyzed by using entropy method;the one-step and two-step transfer probability matrices of visual shift are constructed by Markov theory,and the visual shift characteristics of drivers in normal driving state and different cognitive loads are analyzed;the competitive relationship between visual and cognitive attention under the action of different cognitive loads is summarized,and the driving safety under different cognitive loads is studied from the perspective of vision.The EEG characteristics of the driver were studied.The EEG signals were collected by ICA to remove the eye interference,and the signal was decomposed and reorganized by continuous wavelet transform to obtain five different frequency bands of EEG signals.The amplitude characteristics,significance and signal complexity of the five frequency bands under different driving states were studied,and the EEG topography map was drawn by power spectral density to study the distribution pattern,event-related spectral perturbation and inter-trial coherence index of driver EEG under the effect of cognitive load.The resource occupation pattern of each channel is summarized,and the influence of cognitive load on driving safety is analyzed from the perspective of EEG.Based on EEG indicators to quantify the driver's physiological characteristics index,five EEG indicators are trained by deep learning models to obtain the driver's attention index and relaxation index as the physiological characteristic quantification index.In the construction of the model,the hyperparameters are initially set,and the model is optimized by comparing different activation functions,optimizer optimization algorithms,initial learning rates and the number of hidden layer neurons.To verify the validity of the model,the LSTM network model with the same structure and hyperparameters,the RNN network model and the optimized random forest model were selected for comparison to verify the validity of the model,so as to achieve the quantification of the physiological characteristics of drivers.At the end of this paper,the conflict risk between vehicles and the driving risk induced by the driver's own physiological characteristics are integrated and identified.In the identification method of vehicle conflict risk,the scenario calibration and secondary development of VISSIM are used to match the simulation environment with the traffic environment of the road test,and the multi-vehicle operation data are used to calibrate and calibrate the intelligent driver model and calculate ITTC as the determination index of vehicle conflict risk.In the identification of driving risks caused by drivers' own physiological characteristics,six factors including attention index,relaxation index and pupil diameter are used as risk indicators,and fuzzy Bayesian networks are applied to construct the identification model.In order to effectively improve the safety of driving,two types of risk warning schemes are given at the end of the paper.Effective identification and early warning of driving risks can help reduce potential safety hazards during driving and reduce the number of traffic accidents.It has important practical significance for improving the level of road traffic safety in China and promoting the application of in-vehicle driving assistance systems.At the same time,the research results can also provide policy and theoretical support for traffic managers,and have important application value for strengthening driverrelated management.