Research On Driving Intent Recognition Method Based On Linear Chain Conditional Random Field Model

In recent years,the integration of Intelligent Connected Vehicle(ICV)and Advanced Driver Assistance System(ADAS)has made outstanding contributions in ensuring driving safety,reducing driving burden and improving road traffic efficiency.ADAS provides driving assistance to drivers according to the information of people,vehicles,roads and traffic environment.At present,there is a defect that ADAS can't accurately know the driver's driving intention in the actual working process,and there is a control conflict with the driver.Therefore,the perception and prediction of driver's driving intention is of great significance to avoid control conflict and realize vehicle intelligence.Firstly,Based on linear chain Conditional Random Field(linear-chain CRF),this paper establishes a model for driving intention recognition,and identifies three driving intentions:lane keeping,left lane changing and right lane changing.Then,to provide data support for the training and testing of driving intention recognition model,a static driving simulator is built based on Lab VIEW RT and Car Sim RT to collect data of driving behavior,vehicle state and road environment.The hardware part of static driving simulator is mainly composed of host computer,real-time target computer,steering system,braking/acceleration system and data acquisition card.The real-time target computer is converted from ordinary PC.In addition,NI Veri Stand software is used to realize the functions of real-time I/O channel configuration,data visualization,simulation model and control algorithm import of driving simulator.Car Sim software is used to build vehicle model,road model and driving scene.20 drivers with driving experience were recruited to carry out driving simulation experiment and collect experimental data.Based on the MATLAB programming environment,the box graph detection method is used to detect and eliminate the outliers in the original data.Through data analysis,six kinds of observation parameters were selected to represent driving intention.Finally,based on the driver behavior characteristics and vehicle lateral position data,a personalized driving intention data segmentation method is proposed.In the model learning stage,the data segmentation method is used to establish the training sample database of driving intention recognition model;In the stage of model test,the database of model test samples is established for single condition(lane keeping,left lane changing,right lane changing)and composite condition(left lane changing composite condition,right lane changing composite condition).The model test results show that the driving intention recognition model established in this paper has high accuracy.The research results of this paper can provide a reference for the development of Lane Keeping Assistant System(LKAS)control strategy.In addition,the static driving simulator can be used for the development of vehicle active safety technology,auxiliary driving technology and automatic driving technology.