汽车转向时驾驶员驾驶意图辨识与行为预测

391吉林大学学报(工学版)Vol.39Sup.120093JournalofJilinUniversity(EngineeringandTechnologyEdition)Mar.2009:20080728.:(50775096);(20010576).:(1962-),,,.:.Email:[e-mail]宗长富1,杨肖2,王畅1,张广才3(1.吉林大学汽车动态模拟国家重点实验室,长春130022;2.吉林大学生物与农业工程学院,长春130022;3.辽河油田地球物理勘探公司,辽宁盘锦124010):提出了一种基于隐马尔可夫模型(HMM)和人工神经网络(ANN)相结合的模型,用于实现驾驶意图辨识和驾驶行为预测,从而达到辅助驾驶和提高主动安全性的目的通过驾驶模拟器获得双移线和直线行驶的数据,通过结合HMM和ANN建立了驾驶员意图辨识和行为预测模型,对驾驶员意图和行为在紧急转向正常转向和直线行驶三种工况下的操作分别进行了辨识和预测实验结果表明,该模型在给定车速下能够准确辨识当前的驾驶意图和较为精确预测驾驶员的驾驶行为:车辆工程;隐马尔可夫模型;人工神经网络;驾驶意图辨识;行为预测:U461:A:16715497(2009)Sup.1002706DrivingintentionsidentificationandbehaviorspredictionincarlanechangeZONGChangfu1,YANGXiao2,WANGChang1(1.StatKeyLaboratoryofAutomobileDynamicSimulation,JilinUniversity,Changchun130022,China;2.CollegeofBiologicalandAgriculturalEngineering,JilinUniversity,Changchun130022,China;3.LiaoheFieldGeophysicalProspectingCompany,Panjin124010,China)Abstract:Inordertorealizetheauxiliarydrivingandachievethegoalofimprovingtheactivesafety,anintegratedmodelbasedontheartificialneuralnetworkandthehiddenMarkovchainwasproposedtoidentifyandpredictthedrivingintentionsandbehaviors.Suchamodelwasestablishedwiththedoublelanechangeandstraightlinedataobtainedfromdrivingsimulator.Thismodelwasusedtoidentifyandpredictthedrivingintentionsandbehaviorsunderthisthreecondition,suchasemergencyturning,normalturning,andstraightlinerunning.Experimentalresultsshowthatthemodelcanaccuratelyidentifythecurrentdrivingintentionandaccuratelypredictthenextdrivingbehaviorunderagivenspeed.Keywords:vehicleengineering;hiddenMarkovmodel;artificialneuralnetworks;drivingintentionsidentification;drivingbehaviorprediction,,,ABSTCESPAT(Xbywire),ABSTCESPAT,,Xbywire吉林大学学报(工学版)第39卷,,,,,,[1],,,ToruKumagai[2],TeshengHsiao[3],,()(MM)ANN[4],HMM,HMM,ANN,ANN,ANN,ANNHMM1ANNANN,,,ANN,ANNHMMBayesNet,,[5],,HMM,,,HMM,,HMMANN,HMMANN,ANN,ANN,,HMMANN1.1HMMHMM,HMM,HMMHMM,=[Q,O,aij,bj(k),P]Q={Q1,Q2,!,QT},T=23;O={O1,O2,!,ON}();aij=P(st+1=qj/st=qi)ij,,i,j=1,2,!,N;bj(k)jk;P0,,P(O/),,,BaumWelch,[6],P(O/),ijt(i,j)=t(i)aijbj(Ot+1)!t+1(j)∀Ni=1∀Nj=1t(i)aijbj(xt+1)!t+1(j)(1):at(i)it;!t(i)it;bj(Ot+1)jOt+1BaumWelch^ai,j=∀T-1t=1t(i,j)∀T-1t=1∀t(i)(2)#28#增刊1宗长富,等:汽车转向时驾驶员驾驶意图辨识与行为预测^bi(k)=∀T-1t=1#(ot,vk)∀t(i)∀T-1t=1∀t(i)(3)P^i=∀1(i)(4):^ai,jij;^bj(k)jk;P^i;#(ot,vk)ViterBi[7];∀t(i)tiHMM,ISO[8],90km/h250∃/s,,HMMCE;80km/h250∃/s,,HMMCN;HMMCZ,,:subHMMCX1subHMMCX2subHMMCX3,HMMCX3,O={O1,O2,!,ON}ISOO0,(2)~(4)1,,P(O/1)P(O/),1O[9],,P(O/),,HMMCE,^ai,j=0.00580.80290.19140.00330.38760.60910.98910.00260.0083^bi(k)=!0.0040!2.2641e-21!8.9296e-11!8.6987e-13!5.8243e-23!0.00133%1175P^i=[0.93180.06821.1447e-11]HMMCE,subHMMCEi,HMMCE^ai,j1.2BPBP,[10],BP,,,,BP,BPBP,wmi,bi,wip,bp,SigmoduIi=∀Mm=1wmixmvIi=f∀Mm=1wmixm(5)i=1,2,!,IuPp=∀Ij=1wipvIivPp=f∀Ij=1wipvIi(6)p=1,2,!,P:xm;uIi;vIi;uPp;vPpxm,BPyp=vPp,,Eav,[11]Eav=1N∀Nk=1Ek=12N∀Nk=1∀Pp=1e2kp(7):ekpkp;N,#29#吉林大学学报(工学版)第39卷BPwmi=-0.0003!-0.0299∃∃∃0.0429!0.120212%33wip=89.9423!-90.5925∃∃∃66.6070!-67.410133%12bi=[-0.0008-0.0012!0.0010-0.0008]T1%12bp=[91.104891.2980!9.403267.9531]T1%332HMMANNHMMCEHMMCNHMMCZloglikelihoodHMM,10,,P(O/)3HMM,loglikelihoodHMM,HMM,,subHMMCXiloglikelihood,HMM,(CXi)HMM(CX1CX2!)(CXi+1),ANN,ANN,ANNHMMANN,13[12]5(abcde)90km/h80km/h5254HMMANN,,51HMMANNFig.1CombinationframeworkofHMMandANN(a)=80km/h(b)=90km/h25Fig.2Steeringanglecurvesoffivedriversindoublelaneshiftexamination#30#增刊1宗长富,等:汽车转向时驾驶员驾驶意图辨识与行为预测3.1HMM()3(),1,,HMMCXHMMCX1HMMCXloglikehoodTable1ThreekindsloglikehoodvalueofHMMCXmodelHMMCEHMMCNHMMCZ-49.3653-85.2275-INF-80.0322-59.4357-INF-INF-INF-85.3312,2,subHMMCXi,subHMMCXi2subHMMCXloglikehoodTable2ThreekindsloglikehoodvalueofSUBHMMCXmodelsubHMMCX1subHMMCX2subHMMCX3-39.0902-INF-INF-56.2635-INF-INF,,,(subHMMCX1),HMMCE^ai,j,P1=0.0058P2=0.8029P3=0.1914,3.2BPHMM,,HMM2()HMMBP(),BP,5abcde33Fig.3Comparisonofpredictionvalueandexperimentvalue#31#吉林大学学报(工学版)第39卷,4HMMANNHMM,,ANN,ANN,:[1],,.[J].,2005,22(12):113118.WangYuhai,SongJian,LiXingkun.Studyoninferenceofdriver'sintentionsbasedonfuzzyreasoning[J].JournalofHighwayandTransportationResearchandDevelopment,2005,22(12):113118.[2]KumagaiToru,SakaguchiYasuo,OkuwaMasayuki,etal.Predictionofdrivingbehaviorthroughprobabilisticinference[C]&ProceedingsoftheEighthInternationalConferenceonEngineeringApplicationsofNeuralNetworks(EANN∋03),Torremolinos,Malaga(SPAIN),2003.[3]TeshengHsiao.Timevaryingsystemidentificationviamaximumaposterioriestimationanditsapplicationtodriversteeringmodels[C]&2008AmericanControlConferenceWestinSeattleHotel,Seattle,Washington,USA,2008(6):1113.[4],.BP[J].:,2008,38(3):514519.YangLiya,ShaoChunfu.IntegratedforecastingmodelforrealestatepricealongurbanrailtransitbasedonBPneuralnetworkandMarkovchain[J].JournalofJilinUniversity(EngineeringandTechnologyEdition),2008,38(3):514519.[5]ForbesJ,HuangT,KanazawaK,etal.TheBATmobile:towardsaBayesianautomatedtaxi[C]&The1995InternationalJointConferenceonAI,1995.[6],.[M].:,1988.[7].HMM[D].:,2004.MaGengyu.HMMbasedtemporalsequenceclusteringandrecognition[D].Beijing:QinghuaUniversity,DepartmentofComputerScienceandTechnology,2004.[8]ISO38882.P