Driver Model Considering Visual Cognition Characteristics

As the core decision-making control unit in the closed-loop system of human-vehicle-road,researches on driver modeling are of great significance for designing the decision-making control module of the intelligent driving system.Visual information is the most important source of information for drivers.Studying the characteristics of visual cognition can further analyze the human-vehicle-road interaction relationship and obtain a more humane control effect.Based on an in-depth analysis of domestic and foreign research on driver visual cognition,this paper proposes a driver model considering visual cognition characteristics.Based on the vehicle's current speed and current lane curvature,the model calculates the preview time weight distribution under specific conditions in real-time based on the forward-looking preview time distribution of the actual driver collected in the experiment and uses this as the input of the preview strategy to the driver model based on model predictive control.The following are the specific research contents:First,the driver's visual behaviour during driving is collected through field tests.In the experiment,the driver's visual information,as well as steering behavior information,vehicle kinematics and dynamic information are collected simultaneously.The visual information requires pre-processing including distortion correction,image matching,and the driver gaze point data is obtained through the deviation threshold method(I-DT)from original eye movement data in the visual information.Through JB test(Jarque-Bera test),the normality of the gaze point distribution is verified,and then fit the gaze point data with a general normal distribution,it was found that the mean and standard deviation of the gaze point distribution in linear driving conditions show a significant linear correlation with vehicle speed,The mean value of the distribution of gaze points in curve driving conditions has a significant linear correlation with the logarithm of the curvature of the curve,and the standard deviation of gaze points distribution in curve driving conditions has a significant linear correlation with vehicle speed,witch meansdifferent drivers' visual cognitive characteristics have certain regularity.And the perspective matrix H of the bird's eye projection is calculated by the camera position and pose parameters,the conversion function between the forward-looking preview time and the gaze point is obtained accordingly.Finally,the distribution function of the forward-looking preview time is obtained as a preview strategy for the follow-up driver model.Then,established a driver model based on model predictive control is established and analyzed.The driver model based on model predictive control abstracts the problem of controlling vehicle into an optimization problem with constraints,and further converts the problem into a general QP problem.The prediction model and moving horizon optimization are used to build the driver's predicted behavior systematically,This model has better performance and matches the driver's actual behavior better.Based on this model,the driver's visual cognitive characteristics are integrated,and the preview time distribution under specific conditions is calculated in real-time,which is used as the preview strategy of the driver model.Finally,the driver model considering visual cognition characteristics is verified off-line.The test scenario is a continuous curve,and the test is completed in Car Sim/Simulink platform.The results show that curve tracking performance is significantly improved after considering the visual cognitive characteristics.The lateral tracking error is reduced by a maximum of 30% compared to before visual cognition is integrated,which proves the effectiveness of the proposed modeling method.