Driver Lane Change Intention Recognition And Modeling Based On Driving Behavior And Road Environment

Advanced Driving Assistance System(ADAS)should bring safe and comfortable driving experience to drivers.The ADAS related to lane change mainly includes Lane Change Assist System/Blind Spot Information System(BLIS),Lane Departure Warning System(LDWS)and Lane Keeping Assist System(LKAS).Since these ADASs have defects that do not accurately understand the driver's intentions,they often cause unnecessary interference and false alarms to the driver,resulting in lower utilization.In addition,in the shared control system,it is necessary to obtain the control weight distribution according to the identification results of driver's lane chang intention.Therefore,it is important to identify the driver's lane chang intention accurately.This paper is supported by the National Key Research and Development Plan of China(2016YFB0101102)' Electric Vehicle Intelligent Auxiliary Driving Technology Development and Industrialization' and National Natural Science Foundation(U1664263)"China Automobile Industry Innovation and Development Joint Fund".In this paper,the driver's lane change process is analyzed,the multi-dimension Gauss Hidden Markov Model(MGHMM)of driver intention recognition is established,and the factors affecting the model recognition effect are analyzed,which includs the following aspects:(1)In view of the excellent performance of HMM for time series data modeling,this paper first analyzes the application problem of HMM in driver lane change intention recognition,respectively establishes lane keeping intention recognition HMM model,left lane change intention recognition HMM model and right lane change intention recognition HMM model.In order to improve the recognition performance,based on the HMM driving intention recognition model of single working condition,a lane change intention identification model of composite condition with model matching function is proposed for the personalized driving style.(2)Based on the Carmaker simulation platform,we designed a driving experiment and built a road environment in the Carmaker software in combination with the actual road environment.We analyzed the relationship between various types of state information over time during the lane change process,and collected state information related to straight and lane change driving under different working conditions.The MGHMM observation data is determined by experimental analysis and screening.Based on the parameter characteristics of the MGHMM,we applied statistical theory to determine the method of segmentation of observation data.(3)Through experiments,we analyzed the state and the Gaussian mixture number of the driver's lane change intention recognition MGHMM selection problem,and established the model parameter training database and model verification database under various working conditions.Through the model training and verification process,the influences of model parameters,recognition time window and observation data segmentation on driving intention identification is analyzed.Finally,the significance of establishing the structure of the driver's lane change intention recognition model with composite condition is verified.