| Intelligentization is an important development direction of vehicles in the future.The current research on vehicle intellectualization mainly focuses on the adaptability of traffic environment and decision control,while the research on adaptability of road conditions is less.Road conditions affect driving safety and driving comfort,and are the current research study hotspots.People can sense the road conditions through vision,steering wheel force and bumpy body,and adjust their driving style according to different roads to better complete the driving task.This process shows that people have the ability of self-learning and self-adapting to the road,and intelligent vehicles should also have such abilities to sense the road like people.At present,the research on road identification is mainly focused on the accurate identification of single parameter of road friction coefficient by chassis electronic control system,which cannot meet the needs of intelligent vehicle path planning,decision-making,control and other links.Therefore,it is necessary to conduct research on road type cognition of intelligent vehicle.In order to solve the above problems,this thesis simulates the ability of human drivers who don’t need accurate values of road parameters,and adopts artificial intelligence method to recognize the road environment.On the premise of not adding existing on-board sensors,this thesis selects Gaussian Mixture-Hidden Markov Model as recognition tool,and recognizes the road type corresponding to each wheel in real time according to the vehicle motion response and operation signals obtained from CAN bus.This thesis focuses on the following aspects:First of all,this thesis investigates the existing pattern recognition methods,and focuses on the related training and recognition algorithms of Hidden Markov Models.Secondly,this thesis analyzes the sensitive signals of vehicles for different road features.According to the friction coefficient,rolling resistance coefficient,roughness and slope of the road,the road features are classified comprehensively,and for eachroad attribute,the relevant characteristics and response quantities are analyzed from the theoretical angle,simulation angle and real vehicle angle respectively,thus realizing the selection of sensitive response quantities of the road.Then,using Hidden Markov Model,this thesis studies the identification of road adhesion coefficient and road roughness under ASCL simulation environment.The hardware construction for real-time road identification of real vehicles is completed,and the recognition and verification of road types in real vehicle environment are realized.Finally,the optimization of Hidden Markov Model,data and the technical route have been carried out.Experiments with real vehicle showed that based on the Gaussian Mixture-Hidden Markov Model and dynamic response parameters of vehicles obtained from the CAN bus,the real-time and general cognition of road type corresponding to each wheel can be realized.And the approximate range for the four road parameters--the adhesion coefficient,rolling resistance coefficient,roughness and slope of the road surface--can be obtained. |
