Vision And Dynamics Information Fusion Based Road Adhesion Coefficient Estimation For Intelligent Vehicles

The quality of the road surface is the most direct factor that affects the safe driving of the vehicle,and the road adhesion coefficient represents the maximum interaction force that can be produced between the tire and the road surface,and directly affects the driving,braking and handling stability of the vehicle.Accurately identifying the road adhesion coefficient can expand the range of working conditions of the active safety system of the vehicle,and sensing the changes in the road surface in advance will help the active safety system to adjust the control strategy in time.However,the existing road adhesion coefficient identification methods have a certain degree of estimation error and hysteresis,and it is difficult to deal with sudden changes in road conditions under extreme conditions.Therefore,this article conducts research on how to predict and accurately estimate the road adhesion coefficient in advance.The details are as follows:The accurate estimation of the road adhesion coefficient needs to make full use of the vehicle dynamics response information.This paper adopts the dynamics estimation method based on the Unscented Kalman filter algorithm.A three-degree-of-freedom vehicle dynamics model and a Magic Formula tire model are established,and an Unscented Kalman filter estimator is designed.Subsequently,the estimation system was built on the Car Sim and Matlab/Simulink joint simulation platform.Finally,simulation experiments under different working conditions verified the accuracy of the model and the effectiveness of the estimation algorithm.The estimation error is small and it can cope with sudden changes in the road surface.In order to realize the advance prediction of road conditions,a road type recognition method based on lightweight convolutional neural network is proposed.First,a large number of images of the road ahead were collected through the on-board camera,and eight different road types were obtained through screening and sorting.The data expansion method was used to establish a rich and complete road image information database.In addition,in order to eliminate the impact of non-road areas in the collected images on the classification and recognition tasks,semantic segmentation networks are used to extract the road areas of interest in the images.Finally,the training and evaluation process of the lightweight convolutional neural network is realized through programming,and finally a road type recognition network with high accuracy and high computational efficiency is obtained.Aiming at the estimation of road adhesion coefficient that should meet the requirements of predictability and accuracy,a road adhesion coefficient estimation method that combines visual images and dynamic response information is proposed.A comparison table of road surface types and adhesion coefficient range values under different driving speeds is established,and a spatiotemporal synchronization scheme of image information and dynamic response information is designed.Analyzed and discussed the confidence of the results of the two methods for fusion,and formulated the fusion rules of the estimation system accordingly.Finally,the probability density function truncation method is used to integrate the road image recognition results into the dynamics estimation algorithm in the form of constraints,and the corrected road adhesion coefficient estimation value is obtained.After the detailed design process of the fusion estimation system is given,the Car Sim dynamic simulation software is used to verify the effectiveness of the proposed fusion strategy.The experimental results prove that the fusion strategy can identify the road surface type in advance,predict the adhesion coefficient,and quickly respond to the sudden change of the road surface,and accurately estimate the real-time change of the road adhesion coefficient.