视觉智能控制挡风玻璃雨刮系统的智能汽车发展
视觉智能控制挡风玻璃雨刮系统的智能汽车发展
梓勋朴文镐金弘俊林庆昌和李硕
机械工程釜山国立大学,韩国釜山系
摘要雨刷在降水时对保证驾驶员的安全性起着关键的作用.传统雨刮系统要求司机根据时间和车辆的速度及降水在挡风玻璃上的数量的不断变化,注意调整刮水器的速度和间歇的雨刷间隔.由于雨刮器手动调整会分散驾驶员的注意力,这可能是交通事故的直接原因,许多公司已经开发了采用光学传感器的自动雨刮系统,并取得了不同程度的成功.本文介绍了以视觉为基础的智能雨刷系统,能自动根据挡风玻璃上的水滴量调整其速度和间歇.该系统采用不同的图像处理算法来检测水滴和模糊逻辑来确定速度和刮水器间隔.
关键字智能雨刮系统,智能车辆,雨水感应,雨刷控制,视觉传感器,模糊控制
I.引言
智能车辆提供了提升显著的安全性和方便司机和乘客的潜力.智能车辆是智能交通系统(ITS)的I.个组成部分,其使用各种智能传感和控制算法来评估车辆的环境,并协助司机安全驾驶.这些算法包括驾驶辅助系统,部分控制驱动程序的方便车辆,和碰撞预警系统,让司机避免碰撞提供紧急信息.因为其增强的安全性和便利性能力,智能车辆成为智能交通系统的I.个重要的研究领域.
其中汽车零件影响着驾驶者的安全性和便利性,挡风玻璃刮水器是I.个重要的组成部分,它允许司机降水期间收集视觉信息.然而,传统雨刮系统需要在调整雨刷速度与间歇式雨刮器视觉传感器的间隔,因为在挡风玻璃上的降水数量会根据时间和车辆的速度不断变化而不断变化.因为雨刮器的手动调节会分散驾驶员的注意力,这可能是交通事故的直接原因,许多公司已经开发使用I.些光学传感器有不同程度的成功的自动雨刮系统.
许多智能雨刮系统试图根据雨或雪量自动调节雨刷的速度和间歇区间.这些系统的关键元件是传感器用来测量在挡风玻璃上的水量的.在大多数系统中,光学传感器 *好棒文|www.hbsrm.com +Q: ^3^5`1^9`1^6^0`7^2#
被用于此目的的.这种类型的传感器的使用的事实是,当挡风玻璃是湿的,折射角度和光的反射角度是不同的.许多传感器使用多个LED以小入射角发射光线进入挡风玻璃.光线透过玻璃与几个反射在玻璃和空气边界传播.相对的发光II极管,在几个光传感器的反射后,检测到达的光.当挡风玻璃有水在其表面上时,折射角度与反射光的光量被改变,从而产生不同的光传感器的输出.这种差异是用来确定在挡风玻璃上的水量.
即使光雨传感器被广泛应用,但它具有I.些缺点.I.个缺点是对外部光的灵敏度,这就是为什么许多系统都有I.个黑暗的滤波器对挡风玻璃表面的传感器连接以阻止外界光.然而,当车从隧道或地下停车场出来时许多系统仍然开动雨刮.另I.个缺点,也是I.个主要的缺点,该感测区域是挡风玻璃的I.个相对小的部分.这使得雨刮系统只能用有限的信息进行操作.刮水器系统无法启动时,司机的视线有些雨滴,但不是在检测区.
为了解决这些问题,本文提出了I.种视觉的智能雨刮系统,可以测量I.个比常规系统相对较宽的挡风玻璃面积.尤其是视觉传感器具有的优点,即能测量的不仅有降水量也有雨滴的分布.这些信息可以让系统更加智能化,而不会干扰司机.
在本文中,我们提出了智能雨刮系统的视觉和关键算法的概念.更具体地说,本文展示了I.个有效的雨水感应算法,可以表征白天和黑夜的降雨.此外,本文介绍了I.种模糊逻辑,可以调节转速和区间的智能雨刮控制算法.最后,用实验室环境下的实验结果来评估所提出的系统的有效性.
II视觉的智能雨刷系统的结构
在许多情况下,I.个光学智能雨刷系统的感测区域是非常小的,该区域最宽I.0mm,最长III0mm.这使得整个系统在非常有限的信息下运行.当它开始下雨,雨刮器开始移动后,才有些雨滴落在感应区.此外,当水溅到了其他车辆且超出感应区时,雨刮器无法启动.
为了避免这些问题,我们开发了视觉传感器,如图I.智能雨刷系统.智能雨刮系统由雨刮电机,雨刮器开关,I.个模糊控制器,视觉传感器和发光II极管(LED)组成.这里,刮水器开关用于激活智能雨刷系统.模糊控制器执行两个功能:雨中感知算法,是根据传感器拍摄的降雨强度和视觉图像来计算降雨分布;模糊雨刮器控制算法,是根据降雨强度和分布来计算雨刷速度和雨刮器间隔.最后,将视觉传感器安装在车辆的顶棚捕捉挡风玻璃图像,并且安装LED,以在夜间提供I.些照明.
以保证视觉的智能雨刷系统的性能,车辆的外部背景必须被忽略,并且只有在挡风玻璃上的雨滴可以被由视觉传感器拍摄到的图像提取.为了这个目的,我们通过把光圈开到最大,集中于视觉传感器与现场的深度低的挡风玻璃.最后,该视觉传感器可接受外部光源,如日光,路灯,尾灯和夜间大灯.如太阳光及街灯的部分通过安装视觉传感器向下倾斜来避免干扰,如图I.,但是,其他的干扰,如尾灯和头灯不能被消除.因此,适当的图像处理算法是必要的,以消除这些外部光源的效果.为了这个目的,我们也有对夜间的额外的图像处理步骤.
III智能雨刷系统的雨水感应算法
图II显示了雨水感应算法的视觉的智能雨刷系统.该算法将捕获的图像作为输入,产生沉淀的两个变量,即描述的状态,降雨强度和降雨分布作为其输出.为了获得输出,该算法采用I.系列图像处理技术.在图中,后雨刷开关设置为自动"模式,视觉传感器获取挡风玻璃的图像.然后,根据大灯的打开或关闭,该算法决定额外的夜间处理是否是必须的.经过适当的 *好棒文|www.hbsrm.com +Q: ^3^5`1^9`1^6^0`7^2#
图像处理技术的应用,该算法计算的降雨强度和分布,是用来确定刮水器的速度和时间.降雨强度代表了挡风玻璃上的雨量,雨量分布显示雨滴是如何广泛分布在挡风玻璃上的.
IV模糊雨刮器控制算法
在本文中,模糊控制算法是先调节转速和雨刷间隔的图像处理,然后计算降雨强度和分布.使用模糊逻辑的主要理由是,模糊逻辑可以相对容易地模仿驾驶者的决策,通过调整隶属函数,个别司机可以调整或定制逻辑的结果.
图VI显示了智能雨刷系统的模糊雨刮器控制算法的结构.在图中,模糊控制器由III部分组成:模糊化,推理化和去模糊化.模糊化将降雨强度和分布转换成语言值.推理引擎会使用来自专家经验所产生的模糊控制规则的模糊输出.最后,去模糊化以推断的结果计算雨刮器的速度和雨刷间隔.
图VII显示的模糊输入和输出语言变量的隶属函数.雨量的语言变量被定义为干燥.小雨.雨,而雨分布定义为窄.宽中,如图VII(a)所示.在图中,降雨强度的隶属函数,取决于它是否是白天或夜间.用于刮水器间隔的语言变量被定义为零.短.长,如图VII(b).在这里,我们定义的刮水器间隔从0到III000毫秒变化.转速的语言变量被定义为零.低.高,如图VII(b).
V视觉的智能雨刷系统的性能评价
为了评估视觉的智能雨刮器的功效,我们成立了I.个实验室环境中的实验测试台.该试验台具有从车上拆下的挡风玻璃和雨刷器.在PC机上用图像采集卡(euresysPICOLO板)和CCD相机(PULNIX的TM-II00)实现图像处理和模糊逻辑算法.PC机也连接到I.个接口电路,可以驱动刮水器电机.照相机的光圈调整开放到尽可能宽(F=I..IV).
V.I.白天操作的实验结果
为了模拟白天的毛毛雨,水被喷到挡风玻璃的同时实验室灯的开启.图IX示出了整个处理中的图像,即如前所述的索贝尔掩码.阈值以及扩张.最终的图像中包含III.VIIIIII白色像素(III.VIIIIII降雨强度),I.VI*I.VI单位面积至少有I.个雨滴(III0.V％降雨分布).使用这两个值,根据模糊雨刮器控制算法,我们可以计算出抽头间隔为IIIVV0毫秒,并且抽头的速度是低的.图I.0(a)示出了当大量的雨水下降到白天模拟雨时所捕获的挡风玻璃的图像.从最终的图像看,降雨强度为I.V,VIIIIIII.像素,雨量分布是VIIIIII.II％.这给了我们IIIV0msec雨刷间隔和高雨刷速度.比较这些结果,我们可以看出,虽然III秒钟的最大雨刮器间隔相对于实际应用可能太小了,但模糊逻辑的表现如预期I.样.
VI总结和结论
在本文中,我们提出了I.个视觉的智能雨刮系统,用于提高驾驶的安全性和便利性.该系统采用了I.系列的图像处理步骤,以去除背景的对象和确定在挡风玻璃上的雨滴.从经处理的图像,用模糊逻辑算法选择雨刷速度和时间间隔这两个输入来计算降雨强度和分布.为了证明所提出的系统的有效性,该系统通过I.系列的在实验室设置的实验进行评价.从研究得出的结论如下.
数据表明,通过将适当的图像处理步骤,视觉的雨刮器系统在技术上是可行的.这种视觉的方法比现有的光学系统提供了更多的信息.光学传感器的感测区域可能太小而难以检测稀疏分布的雨滴,且不能直接检测驱动线的雨滴.相比之下,视觉传感器能够覆盖面积足够大来检测视线驾驶员视线分布稀疏的雨滴和雨滴.
降雨强度和分布可能是合适的变量来表征降雨.水的使用量是表征系统在现有的降雨的唯I.变量.然而,我们又增加了变数,用雨水分布来表示雨滴如何分布在挡风玻璃上.通过使用这些变量和I.组简单的模糊规则,我们能够确定雨刷速度和间歇时间间隔.
尽管在上述结论中,由于摄像头,帧捕捉电路,控制器的额外成本,视觉的雨刷系统在经济上还不可行.我们希望,在不久的将来,当智能车辆开始出现,视觉的雨刷系统的优势会大于额外成本和复杂性.当该系统有经济价值时,本文提出的算法应该可以通过观察驾驶员行为的调整在现场和逻辑进行测试.
附件II:外文原文
DevelopmentofVisionbasedControlSmartWindwhieldWiperSystemforIntelligentVehicle
Jee-HunParkMan-HoKimHong-JunIm
Kyung-ChangLeeandSukLee
DepartmentofMechanicalEngineering,PusanNationalUniversity,Busan,Korea
Abstract:Windshieldwipersplayakeyroleinassuringthedriverssafetyduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwiperintervalbecausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.Thispaperpresentsthedevelopmentofvision-basedsmartwindshieldwipersystemthatcanautomaticallyadjustitsspeedandintermittentintervalaccordingtotheamountofwaterdropsonthewindshield.Thesystememploysvariousimageprocessingalgorithmstodetectwaterdropsandfuzzylogictodeterminethespeedandtheintervalofthewiper.
Keywords:SmartWiperSystem,IntelligentVehicle,RainSensing,WiperControl,VisionSensor,FuzzyControl
I..INTRODUCTION
Intelligentvehiclesofferthepotentialtoenhancesafetyandconveniencesignificantlyforbothdriversandpassengers.Asacomponentoftheintelligenttransportationsystem(ITS),theintelligentvehicleusesvariousintelligentsensingandcontrolalgorithmstoassessthevehiclesenvironmentandassistthedriverwithsafedriving.Thesealgorithmsincludethedriverassistancesystemthatpartiallycontrolsthevehicleforthedriversconvenience,andthecollisionwarningsystemthatprovidesemergencyinformationtoallowthedrivertoavoidacollision.Becauseoftheirabilitytoenhancethesafetyandconvenience,intelligentvehiclesbecomeacrucialresearchareafortheintelligenttransportationsystem[I.][II].
Amongthevehiclepartsthatinfluencethedriverssafetyandconvenience,awindshieldwiperisanimportantpartthatallowsadrivertocollectvisualinformationduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwipervisionsensorinterval
becausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.
Manysmartwipersystemstrytoregulatethewipersspeedandintermittentintervalautomaticallyaccordingtotheamountofrainorsnow[III][IV].Thekeyelementofthesesystemsisthesensortomeasuretheamountofwateronthewindshield.Inmostsystems,anopticalsensorisusedforthispurpose[III].Thistypeofsensorsusesthefactthattherefractionangleandtheamountofreflectionofthelightaredifferentwhenthewindshieldiswet.ManysensorsuseseveralLEDstoemitthelightintothewindshieldatasmallincidentangle.Thelighttravelsthroughtheglasswithseveralreflectionsattheglass-airborder.OppositetotheLEDs,thereareseveralphotosensorstodetectthearrivinglightafterthereflections.Whenthewindshieldhaswateronitssurface,therefractionangleandtheamountofreflectedlightarechanged,resultinginthedifferentphotosensoroutput.Thisdifferenceisusedtodeterminetheamountofwateronthewindshield.
Theopticalrainsensorshavesomedisadvantageseventhoughtheyarewidelyused.Oneofdisadvantagesisthesensitivitytoexternallight.Thatiswhymanysystemshaveadarkfilteronthewindshieldsurfacewherethesensorisattachedinordertoblocktheexternallight.However,manysystemsstillactivatethewiperwhenthecarcomesoutoftunnelsorundergroundparkinglot.Anothershortfall,maybeamajoroneisthatthesensingareaisarelativelysmallportionofwindshield.Thismakesthewipersystemoperateonlywithlimitedinformation.Thewipersystemmayfailtoactivatewhentherearesomeraindropsonthedriverslineofsight,butnotonthesensingarea.
Forsolvingtheseproblems,thispaperpresentsavision-basedsmartwipersystemthatcanmeasurearelativelywiderwindshieldareathantheconventionalsystem.Especially,thevisionsensorhasadvantagethatcanmeasurenotonlytheamountofrainbutalsothedistributionofraindrops.Thisinformationcanallowthesystemmoreintelligentlywithoutdisturbingdrivers.
Inthispaper,wepresenttheconceptandkeyalgorithmsforavision-basedsmartwipersystem.Morespecifically,thispapershowsaneffectiverainsensingalgorithmthatcancharacterizetherainfallduringthedayandnight.Inaddition,thispaperintroducesanintelligentwipercontrolalgorithmbasedonfuzzylogicthatcanregulatethewiperspeedandinterval.Finally,experimentalresultsunderthelaboratorysettingsarepresentedtoevaluatetheefficacyoftheproposedsystem.
Thispaperconsistsofsixsectionsincludingthisintroduction.SectionIIintroducesthebasicstructureofavision-basedsmartwipersystem,andSectionIIIpresentstherainsensingalgorithm.TheintelligentwipercontrolsystemusingfuzzylogicisdescribedinSectionIV,andtheexperimentalresultsaregiveninSectionV.Finally,theconclusionsarepresentedinSectionVI.
II.STRUCTUREOFVISION-BASEDSMARTWIPERSYSTEM
Inmanycases,thesensingareaofanopticalsmartwipersystemisverysmall;theareaisatmostI.OmmwideandIII0mmlong.Thismakesthewholesystemoperateonverylimitedinformation.Whenitstartstorain,thewiperbeginstomoveonlyaftersomeraindropsfallontothesensingarea.Moreover,thewipermaynot
startatallwhenwaterissplashedbyothercarsontotheareaotherthanthesensingarea.
Inordertoavoidtheseproblems,wedevelopedasmartwipersystemusingavisionsensorasshowninFig.I..Thesmartwipersystemconsistsofawipermotor,awiperswitch,afuzzycontroller,avisionsensor,andlightemittingdiodes(LEDs).Here,thewiperswitchisusedforactivatingthesmartwipersystem.Thefuzzycontrollerperformstwofunctions:therainsensingalgorithmthatcalculatestherainintensityandraindistributionfromtheimagecapturedbythevisionsensor,andthefuzzywipercontrolalgorithmthatcomputesthewiperspeedandwiperintervalbasedonrainintensityanddistribution.Finally,thevisionsensorisinstalledontheceilingofthevehicletocapturethewindshieldimage,andtheLEDsareinstalledinordertoprovidesomelightingduringthenighttime.
Toguaranteetheperformanceofthevision-basedsmartwipersystem,theexternalbackgroundoutsideofthevehiclemustbeignored,andonlytheraindropsonthewindshieldcanbeextractedfromtheimagecapturedbythevisionsensor.Forthispurpose,wehavefocusedthevisionsensoronthewindshieldwithlowdepthoffieldbyopeningtheaperturetothemaximum.Finally,thevisionsensorcanbeaffectedbyanexternallightsourcesuchassunlightinthedaytime,streetlamps,taillamps,andheadlampsinthenighttime.PartofdisturbancessuchassunlightandstreetlampscanbeavoidedbyinstallingthevisionsensortilteddownwardasshowninFig.I..However,otherdisturbancessuchastaillampsandheadlampscannotbeeliminated.Therefore,anappropriateimageprocessingalgorithmisnecessarytoeliminatetheeffectoftheseexternallightsources.Forthispurpose,wehaveanadditionalimageprocessingstepforthenighttime.
III.RAINSENSINGALGORITHMOFSMARTWIPERSYSTEM
Fig.IIshowstherainsensingalgorithmforthevision-basedsmartwipersystem.Thealgorithmtakesacapturedimageasitsinputandproducesthestateofprecipitationdescribedbytwovariables,i.e.,rainintensityandraindistributionasitsoutputs.Inordertoobtaintheoutputs,thealgorithmemploysaseriesofimageprocessingtechniques.Inthefigure,afterthewiperswitchissetto"auto"mode,thevisionsensoracquiresanimageofthewindshield.Then,dependingonwhethertheheadlampsareonoroff,thealgorithmdecideswhethertheadditionalnighttimeprocessingisrequired.Afterappropriateimageprocessingtechniquesareapplied,thealgorithmcomputesrainintensityanddistributionthatareusedtodeterminethewiperspeedandinterval.Therainintensityrepresentstheamountofrainonthewindshieldwhiletheraindistributionindicateshowwidelytheraindropsaredistributedonthewindshield.
III.I.Daytimeimageprocessingalgorithm
Inordertodetectraindropsonthewindshield,thealgorithmfindstheboundariesofraindropsbyusinganedgedetectiontechnique.AmongmanytechniquessuchasSobel,Prewitt,Roberts,Laplacian,andLaplacianofGaussian[V],theSobelmaskisselectedbecauseitsedgedetectioncapabilityisneithertoobadtodetectraindropsnortoogoodsothattheedgeofbackgroundisdetected.Fig.III(a)showsanoriginalimageofraindropswhileFig.III(b)showstheimageaftertheSobelmaskisapplied.Inthefigure,wecanseethattheboundariesofraindropsareclearwithoutanynoticeableedgesfromthebackground.
TheimageinFig.III(b)stillhastheboundariesofthebackground,thoughnotveryvisible,thatarerepresentedgraypixels.Thisisbecausethebackgroundimageisblurredduetothelowdepthoffield.Inordertoremovethesebackgroundboundariesandshortentheprocessingtime,asinglethresholdTisappliedasshowninEq.(I.)whereandaretheoldandnewvaluesforthepixel,respectively[VI].Inthisequation,thevalueforTisselectedasI.IIVaftersometrialsanderrors.
Fig.III(c)showstheimageafterthresholdingwherewecanseethatthegraybackgroundboundariesarecompletelygone.
Thenextstepistowhitenthepixelsinsidetheboundary.Forthispurpose,weusethedilationoperation[V]asanapproximation.Ingeneral,thedilationmakesanobjectenlargedbecausetheoperationextendstheoutermostpixelsoftheboundary.Therefore,theraindropswillbeslightlylargerandsomeofraindropsstillmayhaveblackpixelsinsidetheboundary.Inthiswork,aIIIXIIIdilationmask,whosepixelvaluesareallzeros,isusedforfastercomputationalspeed[VII].Fig.III(d)showsthefinalimageafterthedilationoperation.ComparingFig.III(c)withFig.III(d),wecanverifythatmostraindropboundariesarefilledwithwhitepixels.
III.IINighttimeimageprocessingalgorithm
Asmentionedearlier,thenighttimeimagestendtocontainmoredisturbancessuchasstreetlamps,taillamps,andheadlamps.Whenthedaytimeprocessingalgorithmisappliedtothenighttimeimage,theeffectofdisturbancesisstillremaininginthefinalimage.Thisindicatesthatthenighttimeprocessingalgorithmshouldbedifferentfromthedaytimealgorithm.
Forthispurpose,wefocusonthefactthatsharpfeatures,ingeneral,haverelativelyhighfrequencyelements,whileblurredfeatureshaverelativelylowfrequencyelements.Thatis,theraindropsundertheinternalLEDlightaredistinctandhavehighfrequencyelementswhereastheexternallightsourcesareblurredbythelowdepthoffieldandhavelowfrequencyelements.Therefore,disturbancesbytheexternallightsourcecanberemovedbyeliminatinglowfrequencyelementsfromtheimage.
Forremovinglowfrequencyelements,wechoosetheButterworthhighpassfilterthatcanbeexpressedasfollows:
WhereD(u,v)isthedistancefromtheoriginto(u,v)onthefrequencyplane.Inthiswork,thecutofffrequencyisselectedtobeIIIII,andnisselectedtobeI.VIviasometrialsanderrors.Fig.IV(a)showsanighttimeimagecontainingataillamp.Fig.IV(b)showstheimagewherethetaillampiseffectivelyremovedafterthehighpassfilterisappliedtotheoriginalimage.
III.IIICharacterizationofprediction
Thewiperoperationalgorithmrequiressomeinputvaluesderivedfromtheprocessedimage.Therefore,weneedastep,i.e.,characterizationofprecipitation,torepresenttheprocessedimagebysomevalues.Inordertoderivethesevalueswerelyonthenotionthatthefactorsthathinderthedriversviewaretheamountofwaterandthenumberofraindropsonthewindshield.First,whenthewindshieldiscoveredbythesignificantamountofwater,thedriverwillhavedifficultyinfindinglanesandobstacles.However,theamountof
waterisnottheonlyfactorbecausethedriversviewisconsiderablyblockedwhennumerousbutverysmallraindropsarecoveringthewindshield.
Therefore,wedefinetwovariables:rainintensityandraindistribution.Therainintensityisdefinedasthenumberofpixelsrepresentingraindrops(itsvalueis
high)intheprocessedimage(IIVVIxIIVVIpixel).Theraindistributioniscomputedasfollows:I.)dividetheoverallareaofIIVVIxIIVVIpixelimageintoI.VIxI.VIpixelunitareas,II)ifthereisatleastonepixelrepresentingaraindropintheI.VIxI.VIpixelunitarea,assignhighvaluetoallpixelsoftheI.VIxI.VIunitarea,III)theraindistributionisdefinedastheratioofthenumberofpixelswithhighvaluetothenumberofoverallpixels(VIV,VIIIVI).
Fig.Vshowsanexamplehowtherainintensityandraindistributionarecalculated.ThelargesquarerepresentstheIIVVIxIIVVIcapturedimagewhilethesmallsquarerepresentstheenlargedI.VIxI.VIunitarea.Fig.V(a)representsthecasefordrizzleormist.SupposethatraindropsareofthesizeofonepixelandtheyareevenlydistributedsothateachI.VIXI.VIunitareacontainsfourraindropsasshowninthesmallsquareinFig.V(a).Then,therainintensityisthenumberofraindroppixelswhichequalsI.0IIIV(IVineachunitareaI.VIxI.VIunitareasintheprocessedimage).TheraindistributionisI.00%becauseallpixelsofaunitareashouldhavehighvalueduetofourpixelsintheunitarea,andallunitareashavehighvalueduetotheevendistribution.Incontrast,Fig.V(b)representsthecaseofasinglebigraindropatthecenteroftheimage.Inthiscase,theraindropisaslargeasfourunitareasthatcontainI.0IIIVpixels.Therefore,therainintensityisI.0IIIV.TheraindistributionisI..VVI%becauseonlyfourunitareashavehighvalueoutofIIVVIunitareas(IVIIIVVIxI.00).
IV.FUZZYWIPERCONTROLALGORITHM
Inthispaper,afuzzycontrolalgorithmisusedtoregulatethewiperspeedandwiperintervalusingthecalculatedrainintensityanddistributionaftertheimageprocessing.Themajorrationaletousefuzzylogicisthatthefuzzylogicallowsmimickingthehumandriversdecisionmakingwithrelativeeaseandthattheoutcomesofthelogiccanbetunedorcustomizedforindividualdriversbyadjustingmembershipfunctions[VIII][IX].
Fig.VIshowsthestructureofthefuzzywipercontrolalgorithmofthesmartwipersystem.Inthefigure,thefuzzycontrollerconsistsofthreeparts:fuzzifier,inferenceengine,anddefuzzifier.Thefuzzifierconvertstherainintensityanddistributionintolinguisticvalues.Theinferenceenginecreatesthefuzzyoutputsusingfuzzycontrolrulesgeneratedfromexpertexperiences.Finally,thedefuzzifiercalculatesthewiperspeedandwiperintervalfromtheinferredresults.
Fig.VIIshowsthemembershipfunctionsofthefuzzyinputandoutputlinguisticvariables.ThelinguisticvariablesfortherainintensityaredefinedasDry,Drizzle,andRainwhilethosefortheraindistributionaredefinedasNarrow,Medium,andWideasshowninFig.VII(a).Inthefigure,themembershipfunctionsofrainintensityaredifferentdependingonwhetheritisdaytimeornighttime.Thisistocompensatetheboundarydetectioncapabilityinthenighttimebecausealmostalltheboundariesofraindropsaredetectedinthedaytimewhilethealgorithmtendstomisssomeraindropsinthenighttime.ThelinguisticvariablesforthewiperintervalaredefinedasZero,Short,andLongasshowninFig.VII(b).Here,wedefinethatthewiperintervalvariesfromzerotoIII,000msec.ThelinguisticvariablesforthewiperspeedaredefinedasZero,Low,andHighasshowninFig.VII(b).
Forfasterexecutionofthefuzzylogiccontroller,theMamdanismin-maxinferencemethodisused.Fordefuzzification,theMoM(Mean-of-Maximum)methodisusedforthewiperspeed,andtheCoA(Center-of-Area)methodisusedforthewiperinterval[I.0][I.I.].ThereasonforusingtheMoMmethodistoobtaintheoutputthat
changesdiscretelyinsteadofvaryingcontinuously.Thisisbecausethewipermotorsoperateattwodifferentspeeds,i.e.,highandlowinsteadofcontinuouslyvaryingspeed.Inthispaper,wehavethewiperstopiftheoutputvalueissmallerthanone,runathighspeediftheoutputvalueislargerthanIX,andrunatlowspeedotherwise.TheoutputmapsforspeedandintervalareshowninFig.VIII.
V.PERFORMANCEEVALUATIONOFVISION-BASEDSMARTWIPERSYSTEM
Inordertoevaluatetheefficacyofthevision-basedsmartwiper,wehavesetupanexperimentaltestbedinalaboratoryenvironment.Thetestbedhasawindshieldandawiperremovedfromacar.Theimageprocessingandfuzzylogicalgorithmsare
implementedonaPCwithaframegrabber(EuresysPicoloboard)andaCCDcamera(PulnixsTM-II00).ThePCisalsoconnectedtoaninterfacecircuitthatcandrivethewipermotor.Thecamerasapertureisadjustedtoopenaswideaspossible(f=I..IV).
V.I.Experimentalresultsfordaytimeoperation
Inordertoemulatethedrizzleinthedaytime,waterissprayedontothewindshieldwiththelaboratorylightsturnedon.Fig.IXshowstheimagesthroughouttheprocessing,i.e.,Sobelmask,thresholding,anddilationasexplainedearlier.ThefinalimagecontainsII,I.VIIIIIIwhitepixels(rainintensityofII,I.VIIIIII)andhasIII0.V%ofI.VI*I.VIunitareaswithatleastoneraindrops(raindistributionofIII0.V%).Usingthesetwovalues,wecancalculatethatthewiperintervalisII,IVV0msec,andthatthewiperspeedislowfromthefuzzywipercontrolalgorithm.Fig.I.0(a)showsthecapturedimageofthewindshieldwhenalargeamountofwaterisdroppedtoemulatetheraininthedaytime.Fromthefinalimage,therainintensityisI.V,VIIIIIII.pixels,andtherain
distributionisVIIIIII.II%.ThisgivesusIIIV0msecforwiperintervalandhighforwiperspeed.Comparingtheseresults,wecanseethatthefuzzylogicbehavesasintendedalthoughthemaximumwiperintervalofthreesecondsmaybetoosmallforrealapplications.
VI.SUMMARYANDCONCLUSION
Inthispaper,weproposedavision-basedsmartwipersystemtoimprovethesafetyandconvenienceofdrivers.Thesystememploysaseriesofimageprocessingstepstoremovebackgroundobjectsandidentifyraindropsonthewindshield.Fromtheprocessedimage,rainintensityanddistributionarecalculatedastwoinputstothe
fuzzylogicalgorithmselectingwiperspeedandinterval.Inordertodemonstratetheefficacyoftheproposedsystem,thesystemwasevaluatedthroughaseriesofexperimentsunderlaboratorysettings.Theconclusionsderivedfromtheresearchareasfollows.
Itisdemonstratedthatthevision-basedwipersystemistechnicallyviablebycombiningappropriateimageprocessingsteps.Thisvision-basedapproachprovidesfarmoreinformationthattheexistingoptics-basedsystems.Theopticalsensorssensingareamaybetoosmalltodetectthesparselydistributedraindropsandthesensorcannotdetecttheraindropsdirectlyonthelineofsightofdriver.Incontrast,thevisionsensorcancoveranarealargeenoughtodetectsparselydistributedraindropsandraindropsonthedriverslineofsight.
Rainintensityanddistributionmaybesuitablevariablestocharacterizearainfall.Theamountofwaterusedtobetheonlyvariabletocharacterizearainfallinexistingsystems.However,wehaveaddedanothervariable,raindistributiontorepresenthowraindropsaredistributedonthewindshield.Byusingthesetovariablesandasetofsimplefuzzyrules,wewereabletodeterminethewiperspeedandintermittentinterval.
Inspiteoftheaboveconclusions,thevision-basedwipersystemisnotstilleconomicallyviableduetotheextracostforcamera,framecapturecircuitry,andcontroller.Wehopethattheadvantagesofvision-basedwipersystemwilloutweightheextracostandcomplexityinthenearfuturewhenintelligentvehiclesbegintoappear.Whentheproposedsystembeginstomakesenseeconomically,theproposedalgorithmsshouldbetestedinthefieldandthelogicsshouldbefine-tunedbyobservingdriversbehavior.
附件II:外文原文
DevelopmentofVisionbasedControlSmartWindwhieldWiperSystemforIntelligentVehicle
Jee-HunParkMan-HoKimHong-JunIm
Kyung-ChangLeeandSukLee
DepartmentofMechanicalEngineering,PusanNationalUniversity,Busan,Korea
Abstract:Windshieldwipersplayakeyroleinassuringthedriverssafetyduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwiperintervalbecausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.Thispaperpresentsthedevelopmentofvision-basedsmartwindshieldwipersystemthatcanautomaticallyadjustitsspeedandintermittentintervalaccordingtotheamountofwaterdropsonthewindshield.Thesystememploysvariousimageprocessingalgorithmstodetectwaterdropsandfuzzylogictodeterminethespeedandtheintervalofthewiper.
Keywords:SmartWiperSystem,IntelligentVehicle,RainSensing,WiperControl,VisionSensor,FuzzyControl
I..INTRODUCTION
Intelligentvehiclesofferthepotentialtoenhancesafetyandconveniencesignificantlyforbothdriversandpassengers.Asacomponentoftheintelligenttransportationsystem(ITS),theintelligentvehicleusesvariousintelligentsensingandcontrolalgorithmstoassessthevehiclesenvironmentandassistthedriverwithsafedriving.Thesealgorithmsincludethedriverassistancesystemthatpartiallycontrolsthevehicleforthedriversconvenience,andthecollisionwarningsystemthatprovidesemergencyinformationtoallowthedrivertoavoidacollision.Becauseoftheirabilitytoenhancethesafetyandconvenience,intelligentvehiclesbecomeacrucialresearchareafortheintelligenttransportationsystem[I.][II].
Amongthevehiclepartsthatinfluencethedriverssafetyandconvenience,awindshieldwiperisanimportantpartthatallowsadrivertocollectvisualinformationduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwipervisionsensorinterval
becausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.
Manysmartwipersystemstrytoregulatethewipersspeedandintermittentintervalautomaticallyaccordingtotheamountofrainorsnow[III][IV].Thekeyelementofthesesystemsisthesensortomeasuretheamountofwateronthewindshield.Inmostsystems,anopticalsensorisusedforthispurpose[III].Thistypeofsensorsusesthefactthattherefractionangleandtheamountofreflectionofthelightaredifferentwhenthewindshieldiswet.ManysensorsuseseveralLEDstoemitthelightintothewindshieldatasmallincidentangle.Thelighttravelsthroughtheglasswithseveralreflectionsattheglass-airborder.OppositetotheLEDs,thereareseveralphotosensorstodetectthearrivinglightafterthereflections.Whenthewindshieldhaswateronitssurface,therefractionangleandtheamountofreflectedlightarechanged,resultinginthedifferentphotosensoroutput.Thisdifferenceisusedtodeterminetheamountofwateronthewindshield.
Theopticalrainsensorshavesomedisadvantageseventhoughtheyarewidelyused.Oneofdisadvantagesisthesensitivitytoexternallight.Thatiswhymanysystemshaveadarkfilteronthewindshieldsurfacewherethesensorisattachedinordertoblocktheexternallight.However,manysystemsstillactivatethewiperwhenthecarcomesoutoftunnelsorundergroundparkinglot.Anothershortfall,maybeamajoroneisthatthesensingareaisarelativelysmallportionofwindshield.Thismakesthewipersystemoperateonlywithlimitedinformation.Thewipersystemmayfailtoactivatewhentherearesomeraindropsonthedriverslineofsight,butnotonthesensingarea.
Forsolvingtheseproblems,thispaperpresentsavision-basedsmartwipersystemthatcanmeasurearelativelywiderwindshieldareathantheconventionalsystem.Especially,thevisionsensorhasadvantagethatcanmeasurenotonlytheamountofrainbutalsothedistributionofraindrops.Thisinformationcanallowthesystemmoreintelligentlywithoutdisturbingdrivers.
Inthispaper,wepresenttheconceptandkeyalgorithmsforavision-basedsmartwipersystem.Morespecifically,thispapershowsaneffectiverainsensingalgorithmthatcancharacterizetherainfallduringthedayandnight.Inaddition,thispaperintroducesanintelligentwipercontrolalgorithmbasedonfuzzylogicthatcanregulatethewiperspeedandinterval.Finally,experimentalresultsunderthelaboratorysettingsarepresentedtoevaluatetheefficacyoftheproposedsystem.
Thispaperconsistsofsixsectionsincludingthisintroduction.SectionIIintroducesthebasicstructureofavision-basedsmartwipersystem,andSectionIIIpresentstherainsensingalgorithm.TheintelligentwipercontrolsystemusingfuzzylogicisdescribedinSectionIV,andtheexperimentalresultsaregiveninSectionV.Finally,theconclusionsarepresentedinSectionVI.
II.STRUCTUREOFVISION-BASEDSMARTWIPERSYSTEM
Inmanycases,thesensingareaofanopticalsmartwipersystemisverysmall;theareaisatmostI.OmmwideandIII0mmlong.Thismakesthewholesystemoperateonverylimitedinformation.Whenitstartstorain,thewiperbeginstomoveonlyaftersomeraindropsfallontothesensingarea.Moreover,thewipermaynot
startatallwhenwaterissplashedbyothercarsontotheareaotherthanthesensingarea.
Inordertoavoidtheseproblems,wedevelopedasmartwipersystemusingavisionsensorasshowninFig.I..Thesmartwipersystemconsistsofawipermotor,awiperswitch,afuzzycontroller,avisionsensor,andlightemittingdiodes(LEDs).Here,thewiperswitchisusedforactivatingthesmartwipersystem.Thefuzzycontrollerperformstwofunctions:therainsensingalgorithmthatcalculatestherainintensityandraindistributionfromtheimagecapturedbythevisionsensor,andthefuzzywipercontrolalgorithmthatcomputesthewiperspeedandwiperintervalbasedonrainintensityanddistribution.Finally,thevisionsensorisinstalledontheceilingofthevehicletocapturethewindshieldimage,andtheLEDsareinstalledinordertoprovidesomelightingduringthenighttime.
Toguaranteetheperformanceofthevision-basedsmartwipersystem,theexternalbackgroundoutsideofthevehiclemustbeignored,andonlytheraindropsonthewindshieldcanbeextractedfromtheimagecapturedbythevisionsensor.Forthispurpose,wehavefocusedthevisionsensoronthewindshieldwithlowdepthoffieldbyopeningtheaperturetothemaximum.Finally,thevisionsensorcanbeaffectedbyanexternallightsourcesuchassunlightinthedaytime,streetlamps,taillamps,andheadlampsinthenighttime.PartofdisturbancessuchassunlightandstreetlampscanbeavoidedbyinstallingthevisionsensortilteddownwardasshowninFig.I..However,otherdisturbancessuchastaillampsandheadlampscannotbeeliminated.Therefore,anappropriateimageprocessingalgorithmisnecessarytoeliminatetheeffectoftheseexternallightsources.Forthispurpose,wehaveanadditionalimageprocessingstepforthenighttime.
III.RAINSENSINGALGORITHMOFSMARTWIPERSYSTEM
Fig.IIshowstherainsensingalgorithmforthevision-basedsmartwipersystem.Thealgorithmtakesacapturedimageasitsinputandproducesthestateofprecipitationdescribedbytwovariables,i.e.,rainintensityandraindistributionasitsoutputs.Inordertoobtaintheoutputs,thealgorithmemploysaseriesofimageprocessingtechniques.Inthefigure,afterthewiperswitchissetto"auto"mode,thevisionsensoracquiresanimageofthewindshield.Then,dependingonwhethertheheadlampsareonoroff,thealgorithmdecideswhethertheadditionalnighttimeprocessingisrequired.Afterappropriateimageprocessingtechniquesareapplied,thealgorithmcomputesrainintensityanddistributionthatareusedtodeterminethewiperspeedandinterval.Therainintensityrepresentstheamountofrainonthewindshieldwhiletheraindistributionindicateshowwidelytheraindropsaredistributedonthewindshield.
III.I.Daytimeimageprocessingalgorithm
Inordertodetectraindropsonthewindshield,thealgorithmfindstheboundariesofraindropsbyusinganedgedetectiontechnique.AmongmanytechniquessuchasSobel,Prewitt,Roberts,Laplacian,andLaplacianofGaussian[V],theSobelmaskisselectedbecauseitsedgedetectioncapabilityisneithertoobadtodetectraindropsnortoogoodsothattheedgeofbackgroundisdetected.Fig.III(a)showsanoriginalimageofraindropswhileFig.III(b)showstheimageaftertheSobelmaskisapplied.Inthefigure,wecanseethattheboundariesofraindropsareclearwithoutanynoticeableedgesfromthebackground.
TheimageinFig.III(b)stillhastheboundariesofthebackground,thoughnotveryvisible,thatarerepresentedgraypixels.Thisisbecausethebackgroundimageisblurredduetothelowdepthoffield.Inordertoremovethesebackgroundboundariesandshortentheprocessingtime,asinglethresholdTisappliedasshowninEq.(I.)whereandaretheoldandnewvaluesforthepixel,respectively[VI].Inthisequation,thevalueforTisselectedasI.IIVaftersometrialsanderrors.
Fig.III(c)showstheimageafterthresholdingwherewecanseethatthegraybackgroundboundariesarecompletelygone.
Thenextstepistowhitenthepixelsinsidetheboundary.Forthispurpose,weusethedilationoperation[V]asanapproximation.Ingeneral,thedilationmakesanobjectenlargedbecausetheoperationextendstheoutermostpixelsoftheboundary.Therefore,theraindropswillbeslightlylargerandsomeofraindropsstillmayhaveblackpixelsinsidetheboundary.Inthiswork,aIIIXIIIdilationmask,whosepixelvaluesareallzeros,isusedforfastercomputationalspeed[VII].Fig.III(d)showsthefinalimageafterthedilationoperation.ComparingFig.III(c)withFig.III(d),wecanverifythatmostraindropboundariesarefilledwithwhitepixels.
III.IINighttimeimageprocessingalgorithm
Asmentionedearlier,thenighttimeimagestendtocontainmoredisturbancessuchasstreetlamps,taillamps,andheadlamps.Whenthedaytimeprocessingalgorithmisappliedtothenighttimeimage,theeffectofdisturbancesisstillremaininginthefinalimage.Thisindicatesthatthenighttimeprocessingalgorithmshouldbedifferentfromthedaytimealgorithm.
Forthispurpose,wefocusonthefactthatsharpfeatures,ingeneral,haverelativelyhighfrequencyelements,whileblurredfeatureshaverelativelylowfrequencyelements.Thatis,theraindropsundertheinternalLEDlightaredistinctandhavehighfrequencyelementswhereastheexternallightsourcesareblurredbythelowdepthoffieldandhavelowfrequencyelements.Therefore,disturbancesbytheexternallightsourcecanberemovedbyeliminatinglowfrequencyelementsfromtheimage.
Forremovinglowfrequencyelements,wechoosetheButterworthhighpassfilterthatcanbeexpressedasfollows:
WhereD(u,v)isthedistancefromtheoriginto(u,v)onthefrequencyplane.Inthiswork,thecutofffrequencyisselectedtobeIIIII,andnisselectedtobeI.VIviasometrialsanderrors.Fig.IV(a)showsanighttimeimagecontainingataillamp.Fig.IV(b)showstheimagewherethetaillampiseffectivelyremovedafterthehighpassfilterisappliedtotheoriginalimage.
III.IIICharacterizationofprediction
Thewiperoperationalgorithmrequiressomeinputvaluesderivedfromtheprocessedimage.Therefore,weneedastep,i.e.,characterizationofprecipitation,torepresenttheprocessedimagebysomevalues.Inordertoderivethesevalueswerelyonthenotionthatthefactorsthathinderthedriversviewaretheamountofwaterandthenumberofraindropsonthewindshield.First,whenthewindshieldiscoveredbythesignificantamountofwater,thedriverwillhavedifficultyinfindinglanesandobstacles.However,theamountof
waterisnottheonlyfactorbecausethedriversviewisconsiderablyblockedwhennumerousbutverysmallraindropsarecoveringthewindshield.
Therefore,wedefinetwovariables:rainintensityandraindistribution.Therainintensityisdefinedasthenumberofpixelsrepresentingraindrops(itsvalueis
high)intheprocessedimage(IIVVIxIIVVIpixel).Theraindistributioniscomputedasfollows:I.)dividetheoverallareaofIIVVIxIIVVIpixelimageintoI.VIxI.VIpixelunitareas,II)ifthereisatleastonepixelrepresentingaraindropintheI.VIxI.VIpixelunitarea,assignhighvaluetoallpixelsoftheI.VIxI.VIunitarea,III)theraindistributionisdefinedastheratioofthenumberofpixelswithhighvaluetothenumberofoverallpixels(VIV,VIIIVI).
Fig.Vshowsanexamplehowtherainintensityandraindistributionarecalculated.ThelargesquarerepresentstheIIVVIxIIVVIcapturedimagewhilethesmallsquarerepresentstheenlargedI.VIxI.VIunitarea.Fig.V(a)representsthecasefordrizzleormist.SupposethatraindropsareofthesizeofonepixelandtheyareevenlydistributedsothateachI.VIXI.VIunitareacontainsfourraindropsasshowninthesmallsquareinFig.V(a).Then,therainintensityisthenumberofraindroppixelswhichequalsI.0IIIV(IVineachunitareaI.VIxI.VIunitareasintheprocessedimage).TheraindistributionisI.00%becauseallpixelsofaunitareashouldhavehighvalueduetofourpixelsintheunitarea,andallunitareashavehighvalueduetotheevendistribution.Incontrast,Fig.V(b)representsthecaseofasinglebigraindropatthecenteroftheimage.Inthiscase,theraindropisaslargeasfourunitareasthatcontainI.0IIIVpixels.Therefore,therainintensityisI.0IIIV.TheraindistributionisI..VVI%becauseonlyfourunitareashavehighvalueoutofIIVVIunitareas(IVIIIVVIxI.00).
IV.FUZZYWIPERCONTROLALGORITHM
Inthispaper,afuzzycontrolalgorithmisusedtoregulatethewiperspeedandwiperintervalusingthecalculatedrainintensityanddistributionaftertheimageprocessing.Themajorrationaletousefuzzylogicisthatthefuzzylogicallowsmimickingthehumandriversdecisionmakingwithrelativeeaseandthattheoutcomesofthelogiccanbetunedorcustomizedforindividualdriversbyadjustingmembershipfunctions[VIII][IX].
Fig.VIshowsthestructureofthefuzzywipercontrolalgorithmofthesmartwipersystem.Inthefigure,thefuzzycontrollerconsistsofthreeparts:fuzzifier,inferenceengine,anddefuzzifier.Thefuzzifierconvertstherainintensityanddistributionintolinguisticvalues.Theinferenceenginecreatesthefuzzyoutputsusingfuzzycontrolrulesgeneratedfromexpertexperiences.Finally,thedefuzzifiercalculatesthewiperspeedandwiperintervalfromtheinferredresults.
Fig.VIIshowsthemembershipfunctionsofthefuzzyinputandoutputlinguisticvariables.ThelinguisticvariablesfortherainintensityaredefinedasDry,Drizzle,andRainwhilethosefortheraindistributionaredefinedasNarrow,Medium,andWideasshowninFig.VII(a).Inthefigure,themembershipfunctionsofrainintensityaredifferentdependingonwhetheritisdaytimeornighttime.Thisistocompensatetheboundarydetectioncapabilityinthenighttimebecausealmostalltheboundariesofraindropsaredetectedinthedaytimewhilethealgorithmtendstomisssomeraindropsinthenighttime.ThelinguisticvariablesforthewiperintervalaredefinedasZero,Short,andLongasshowninFig.VII(b).Here,wedefinethatthewiperintervalvariesfromzerotoIII,000msec.ThelinguisticvariablesforthewiperspeedaredefinedasZero,Low,andHighasshowninFig.VII(b).
Forfasterexecutionofthefuzzylogiccontroller,theMamdanismin-maxinferencemethodisused.Fordefuzzification,theMoM(Mean-of-Maximum)methodisusedforthewiperspeed,andtheCoA(Center-of-Area)methodisusedforthewiperinterval[I.0][I.I.].ThereasonforusingtheMoMmethodistoobtaintheoutputthat
changesdiscretelyinsteadofvaryingcontinuously.Thisisbecausethewipermotorsoperateattwodifferentspeeds,i.e.,highandlowinsteadofcontinuouslyvaryingspeed.Inthispaper,wehavethewiperstopiftheoutputvalueissmallerthanone,runathighspeediftheoutputvalueislargerthanIX,andrunatlowspeedotherwise.TheoutputmapsforspeedandintervalareshowninFig.VIII.
V.PERFORMANCEEVALUATIONOFVISION-BASEDSMARTWIPERSYSTEM
Inordertoevaluatetheefficacyofthevision-basedsmartwiper,wehavesetupanexperimentaltestbedinalaboratoryenvironment.Thetestbedhasawindshieldandawiperremovedfromacar.Theimageprocessingandfuzzylogicalgorithmsare
implementedonaPCwithaframegrabber(EuresysPicoloboard)andaCCDcamera(PulnixsTM-II00).ThePCisalsoconnectedtoaninterfacecircuitthatcandrivethewipermotor.Thecamerasapertureisadjustedtoopenaswideaspossible(f=I..IV).
V.I.Experimentalresultsfordaytimeoperation
Inordertoemulatethedrizzleinthedaytime,waterissprayedontothewindshieldwiththelaboratorylightsturnedon.Fig.IXshowstheimagesthroughouttheprocessing,i.e.,Sobelmask,thresholding,anddilationasexplainedearlier.ThefinalimagecontainsII,I.VIIIIIIwhitepixels(rainintensityofII,I.VIIIIII)andhasIII0.V%ofI.VI*I.VIunitareaswithatleastoneraindrops(raindistributionofIII0.V%).Usingthesetwovalues,wecancalculatethatthewiperintervalisII,IVV0msec,andthatthewiperspeedislowfromthefuzzywipercontrolalgorithm.Fig.I.0(a)showsthecapturedimageofthewindshieldwhenalargeamountofwaterisdroppedtoemulatetheraininthedaytime.Fromthefinalimage,therainintensityisI.V,VIIIIIII.pixels,andtherain
distributionisVIIIIII.II%.ThisgivesusIIIV0msecforwiperintervalandhighforwiperspeed.Comparingtheseresults,wecanseethatthefuzzylogicbehavesasintendedalthoughthemaximumwiperintervalofthreesecondsmaybetoosmallforrealapplications.
VI.SUMMARYANDCONCLUSION
Inthispaper,weproposedavision-basedsmartwipersystemtoimprovethesafetyandconvenienceofdrivers.Thesystememploysaseriesofimageprocessingstepstoremovebackgroundobjectsandidentifyraindropsonthewindshield.Fromtheprocessedimage,rainintensityanddistributionarecalculatedastwoinputstothe
fuzzylogicalgorithmselectingwiperspeedandinterval.Inordertodemonstratetheefficacyoftheproposedsystem,thesystemwasevaluatedthroughaseriesofexperimentsunderlaboratorysettings.Theconclusionsderivedfromtheresearchareasfollows.
Itisdemonstratedthatthevision-basedwipersystemistechnicallyviablebycombiningappropriateimageprocessingsteps.Thisvision-basedapproachprovidesfarmoreinformationthattheexistingoptics-basedsystems.Theopticalsensorssensingareamaybetoosmalltodetectthesparselydistributedraindropsandthesensorcannotdetecttheraindropsdirectlyonthelineofsightofdriver.Incontrast,thevisionsensorcancoveranarealargeenoughtodetectsparselydistributedraindropsandraindropsonthedriverslineofsight.
Rainintensityanddistributionmaybesuitablevariablestocharacterizearainfall.Theamountofwaterusedtobetheonlyvariabletocharacterizearainfallinexistingsystems.However,wehaveaddedanothervariable,raindistributiontorepresenthowraindropsaredistributedonthewindshield.Byusingthesetovariablesandasetofsimplefuzzyrules,wewereabletodeterminethewiperspeedandintermittentinterval.
Inspiteoftheaboveconclusions,thevision-basedwipersystemisnotstilleconomicallyviableduetotheextracostforcamera,framecapturecircuitry,andcontroller.Wehopethattheadvantagesofvision-basedwipersystemwilloutweightheextracostandcomplexityinthenearfuturewhenintelligentvehiclesbegintoappear.Whentheproposedsystembeginstomakesenseeconomically,theproposedalgorithmsshouldbetestedinthefieldandthelogicsshouldbefine-tunedbyobservingdriversbehavior.
梓勋朴文镐金弘俊林庆昌和李硕
机械工程釜山国立大学,韩国釜山系
摘要雨刷在降水时对保证驾驶员的安全性起着关键的作用.传统雨刮系统要求司机根据时间和车辆的速度及降水在挡风玻璃上的数量的不断变化,注意调整刮水器的速度和间歇的雨刷间隔.由于雨刮器手动调整会分散驾驶员的注意力,这可能是交通事故的直接原因,许多公司已经开发了采用光学传感器的自动雨刮系统,并取得了不同程度的成功.本文介绍了以视觉为基础的智能雨刷系统,能自动根据挡风玻璃上的水滴量调整其速度和间歇.该系统采用不同的图像处理算法来检测水滴和模糊逻辑来确定速度和刮水器间隔.
关键字智能雨刮系统,智能车辆,雨水感应,雨刷控制,视觉传感器,模糊控制
I.引言
智能车辆提供了提升显著的安全性和方便司机和乘客的潜力.智能车辆是智能交通系统(ITS)的I.个组成部分,其使用各种智能传感和控制算法来评估车辆的环境,并协助司机安全驾驶.这些算法包括驾驶辅助系统,部分控制驱动程序的方便车辆,和碰撞预警系统,让司机避免碰撞提供紧急信息.因为其增强的安全性和便利性能力,智能车辆成为智能交通系统的I.个重要的研究领域.
其中汽车零件影响着驾驶者的安全性和便利性,挡风玻璃刮水器是I.个重要的组成部分,它允许司机降水期间收集视觉信息.然而,传统雨刮系统需要在调整雨刷速度与间歇式雨刮器视觉传感器的间隔,因为在挡风玻璃上的降水数量会根据时间和车辆的速度不断变化而不断变化.因为雨刮器的手动调节会分散驾驶员的注意力,这可能是交通事故的直接原因,许多公司已经开发使用I.些光学传感器有不同程度的成功的自动雨刮系统.
许多智能雨刮系统试图根据雨或雪量自动调节雨刷的速度和间歇区间.这些系统的关键元件是传感器用来测量在挡风玻璃上的水量的.在大多数系统中,光学传感器 *好棒文|www.hbsrm.com +Q: ^3^5`1^9`1^6^0`7^2#
被用于此目的的.这种类型的传感器的使用的事实是,当挡风玻璃是湿的,折射角度和光的反射角度是不同的.许多传感器使用多个LED以小入射角发射光线进入挡风玻璃.光线透过玻璃与几个反射在玻璃和空气边界传播.相对的发光II极管,在几个光传感器的反射后,检测到达的光.当挡风玻璃有水在其表面上时,折射角度与反射光的光量被改变,从而产生不同的光传感器的输出.这种差异是用来确定在挡风玻璃上的水量.
即使光雨传感器被广泛应用,但它具有I.些缺点.I.个缺点是对外部光的灵敏度,这就是为什么许多系统都有I.个黑暗的滤波器对挡风玻璃表面的传感器连接以阻止外界光.然而,当车从隧道或地下停车场出来时许多系统仍然开动雨刮.另I.个缺点,也是I.个主要的缺点,该感测区域是挡风玻璃的I.个相对小的部分.这使得雨刮系统只能用有限的信息进行操作.刮水器系统无法启动时,司机的视线有些雨滴,但不是在检测区.
为了解决这些问题,本文提出了I.种视觉的智能雨刮系统,可以测量I.个比常规系统相对较宽的挡风玻璃面积.尤其是视觉传感器具有的优点,即能测量的不仅有降水量也有雨滴的分布.这些信息可以让系统更加智能化,而不会干扰司机.
在本文中,我们提出了智能雨刮系统的视觉和关键算法的概念.更具体地说,本文展示了I.个有效的雨水感应算法,可以表征白天和黑夜的降雨.此外,本文介绍了I.种模糊逻辑,可以调节转速和区间的智能雨刮控制算法.最后,用实验室环境下的实验结果来评估所提出的系统的有效性.
II视觉的智能雨刷系统的结构
在许多情况下,I.个光学智能雨刷系统的感测区域是非常小的,该区域最宽I.0mm,最长III0mm.这使得整个系统在非常有限的信息下运行.当它开始下雨,雨刮器开始移动后,才有些雨滴落在感应区.此外,当水溅到了其他车辆且超出感应区时,雨刮器无法启动.
为了避免这些问题,我们开发了视觉传感器,如图I.智能雨刷系统.智能雨刮系统由雨刮电机,雨刮器开关,I.个模糊控制器,视觉传感器和发光II极管(LED)组成.这里,刮水器开关用于激活智能雨刷系统.模糊控制器执行两个功能:雨中感知算法,是根据传感器拍摄的降雨强度和视觉图像来计算降雨分布;模糊雨刮器控制算法,是根据降雨强度和分布来计算雨刷速度和雨刮器间隔.最后,将视觉传感器安装在车辆的顶棚捕捉挡风玻璃图像,并且安装LED,以在夜间提供I.些照明.
以保证视觉的智能雨刷系统的性能,车辆的外部背景必须被忽略,并且只有在挡风玻璃上的雨滴可以被由视觉传感器拍摄到的图像提取.为了这个目的,我们通过把光圈开到最大,集中于视觉传感器与现场的深度低的挡风玻璃.最后,该视觉传感器可接受外部光源,如日光,路灯,尾灯和夜间大灯.如太阳光及街灯的部分通过安装视觉传感器向下倾斜来避免干扰,如图I.,但是,其他的干扰,如尾灯和头灯不能被消除.因此,适当的图像处理算法是必要的,以消除这些外部光源的效果.为了这个目的,我们也有对夜间的额外的图像处理步骤.
III智能雨刷系统的雨水感应算法
图II显示了雨水感应算法的视觉的智能雨刷系统.该算法将捕获的图像作为输入,产生沉淀的两个变量,即描述的状态,降雨强度和降雨分布作为其输出.为了获得输出,该算法采用I.系列图像处理技术.在图中,后雨刷开关设置为自动"模式,视觉传感器获取挡风玻璃的图像.然后,根据大灯的打开或关闭,该算法决定额外的夜间处理是否是必须的.经过适当的 *好棒文|www.hbsrm.com +Q: ^3^5`1^9`1^6^0`7^2#
图像处理技术的应用,该算法计算的降雨强度和分布,是用来确定刮水器的速度和时间.降雨强度代表了挡风玻璃上的雨量,雨量分布显示雨滴是如何广泛分布在挡风玻璃上的.
IV模糊雨刮器控制算法
在本文中,模糊控制算法是先调节转速和雨刷间隔的图像处理,然后计算降雨强度和分布.使用模糊逻辑的主要理由是,模糊逻辑可以相对容易地模仿驾驶者的决策,通过调整隶属函数,个别司机可以调整或定制逻辑的结果.
图VI显示了智能雨刷系统的模糊雨刮器控制算法的结构.在图中,模糊控制器由III部分组成:模糊化,推理化和去模糊化.模糊化将降雨强度和分布转换成语言值.推理引擎会使用来自专家经验所产生的模糊控制规则的模糊输出.最后,去模糊化以推断的结果计算雨刮器的速度和雨刷间隔.
图VII显示的模糊输入和输出语言变量的隶属函数.雨量的语言变量被定义为干燥.小雨.雨,而雨分布定义为窄.宽中,如图VII(a)所示.在图中,降雨强度的隶属函数,取决于它是否是白天或夜间.用于刮水器间隔的语言变量被定义为零.短.长,如图VII(b).在这里,我们定义的刮水器间隔从0到III000毫秒变化.转速的语言变量被定义为零.低.高,如图VII(b).
V视觉的智能雨刷系统的性能评价
为了评估视觉的智能雨刮器的功效,我们成立了I.个实验室环境中的实验测试台.该试验台具有从车上拆下的挡风玻璃和雨刷器.在PC机上用图像采集卡(euresysPICOLO板)和CCD相机(PULNIX的TM-II00)实现图像处理和模糊逻辑算法.PC机也连接到I.个接口电路,可以驱动刮水器电机.照相机的光圈调整开放到尽可能宽(F=I..IV).
V.I.白天操作的实验结果
为了模拟白天的毛毛雨,水被喷到挡风玻璃的同时实验室灯的开启.图IX示出了整个处理中的图像,即如前所述的索贝尔掩码.阈值以及扩张.最终的图像中包含III.VIIIIII白色像素(III.VIIIIII降雨强度),I.VI*I.VI单位面积至少有I.个雨滴(III0.V％降雨分布).使用这两个值,根据模糊雨刮器控制算法,我们可以计算出抽头间隔为IIIVV0毫秒,并且抽头的速度是低的.图I.0(a)示出了当大量的雨水下降到白天模拟雨时所捕获的挡风玻璃的图像.从最终的图像看,降雨强度为I.V,VIIIIIII.像素,雨量分布是VIIIIII.II％.这给了我们IIIV0msec雨刷间隔和高雨刷速度.比较这些结果,我们可以看出,虽然III秒钟的最大雨刮器间隔相对于实际应用可能太小了,但模糊逻辑的表现如预期I.样.
VI总结和结论
在本文中,我们提出了I.个视觉的智能雨刮系统,用于提高驾驶的安全性和便利性.该系统采用了I.系列的图像处理步骤,以去除背景的对象和确定在挡风玻璃上的雨滴.从经处理的图像,用模糊逻辑算法选择雨刷速度和时间间隔这两个输入来计算降雨强度和分布.为了证明所提出的系统的有效性,该系统通过I.系列的在实验室设置的实验进行评价.从研究得出的结论如下.
数据表明,通过将适当的图像处理步骤,视觉的雨刮器系统在技术上是可行的.这种视觉的方法比现有的光学系统提供了更多的信息.光学传感器的感测区域可能太小而难以检测稀疏分布的雨滴,且不能直接检测驱动线的雨滴.相比之下,视觉传感器能够覆盖面积足够大来检测视线驾驶员视线分布稀疏的雨滴和雨滴.
降雨强度和分布可能是合适的变量来表征降雨.水的使用量是表征系统在现有的降雨的唯I.变量.然而,我们又增加了变数,用雨水分布来表示雨滴如何分布在挡风玻璃上.通过使用这些变量和I.组简单的模糊规则,我们能够确定雨刷速度和间歇时间间隔.
尽管在上述结论中,由于摄像头,帧捕捉电路,控制器的额外成本,视觉的雨刷系统在经济上还不可行.我们希望,在不久的将来,当智能车辆开始出现,视觉的雨刷系统的优势会大于额外成本和复杂性.当该系统有经济价值时,本文提出的算法应该可以通过观察驾驶员行为的调整在现场和逻辑进行测试.
附件II:外文原文
DevelopmentofVisionbasedControlSmartWindwhieldWiperSystemforIntelligentVehicle
Jee-HunParkMan-HoKimHong-JunIm
Kyung-ChangLeeandSukLee
DepartmentofMechanicalEngineering,PusanNationalUniversity,Busan,Korea
Abstract:Windshieldwipersplayakeyroleinassuringthedriverssafetyduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwiperintervalbecausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.Thispaperpresentsthedevelopmentofvision-basedsmartwindshieldwipersystemthatcanautomaticallyadjustitsspeedandintermittentintervalaccordingtotheamountofwaterdropsonthewindshield.Thesystememploysvariousimageprocessingalgorithmstodetectwaterdropsandfuzzylogictodeterminethespeedandtheintervalofthewiper.
Keywords:SmartWiperSystem,IntelligentVehicle,RainSensing,WiperControl,VisionSensor,FuzzyControl
I..INTRODUCTION
Intelligentvehiclesofferthepotentialtoenhancesafetyandconveniencesignificantlyforbothdriversandpassengers.Asacomponentoftheintelligenttransportationsystem(ITS),theintelligentvehicleusesvariousintelligentsensingandcontrolalgorithmstoassessthevehiclesenvironmentandassistthedriverwithsafedriving.Thesealgorithmsincludethedriverassistancesystemthatpartiallycontrolsthevehicleforthedriversconvenience,andthecollisionwarningsystemthatprovidesemergencyinformationtoallowthedrivertoavoidacollision.Becauseoftheirabilitytoenhancethesafetyandconvenience,intelligentvehiclesbecomeacrucialresearchareafortheintelligenttransportationsystem[I.][II].
Amongthevehiclepartsthatinfluencethedriverssafetyandconvenience,awindshieldwiperisanimportantpartthatallowsadrivertocollectvisualinformationduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwipervisionsensorinterval
becausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.
Manysmartwipersystemstrytoregulatethewipersspeedandintermittentintervalautomaticallyaccordingtotheamountofrainorsnow[III][IV].Thekeyelementofthesesystemsisthesensortomeasuretheamountofwateronthewindshield.Inmostsystems,anopticalsensorisusedforthispurpose[III].Thistypeofsensorsusesthefactthattherefractionangleandtheamountofreflectionofthelightaredifferentwhenthewindshieldiswet.ManysensorsuseseveralLEDstoemitthelightintothewindshieldatasmallincidentangle.Thelighttravelsthroughtheglasswithseveralreflectionsattheglass-airborder.OppositetotheLEDs,thereareseveralphotosensorstodetectthearrivinglightafterthereflections.Whenthewindshieldhaswateronitssurface,therefractionangleandtheamountofreflectedlightarechanged,resultinginthedifferentphotosensoroutput.Thisdifferenceisusedtodeterminetheamountofwateronthewindshield.
Theopticalrainsensorshavesomedisadvantageseventhoughtheyarewidelyused.Oneofdisadvantagesisthesensitivitytoexternallight.Thatiswhymanysystemshaveadarkfilteronthewindshieldsurfacewherethesensorisattachedinordertoblocktheexternallight.However,manysystemsstillactivatethewiperwhenthecarcomesoutoftunnelsorundergroundparkinglot.Anothershortfall,maybeamajoroneisthatthesensingareaisarelativelysmallportionofwindshield.Thismakesthewipersystemoperateonlywithlimitedinformation.Thewipersystemmayfailtoactivatewhentherearesomeraindropsonthedriverslineofsight,butnotonthesensingarea.
Forsolvingtheseproblems,thispaperpresentsavision-basedsmartwipersystemthatcanmeasurearelativelywiderwindshieldareathantheconventionalsystem.Especially,thevisionsensorhasadvantagethatcanmeasurenotonlytheamountofrainbutalsothedistributionofraindrops.Thisinformationcanallowthesystemmoreintelligentlywithoutdisturbingdrivers.
Inthispaper,wepresenttheconceptandkeyalgorithmsforavision-basedsmartwipersystem.Morespecifically,thispapershowsaneffectiverainsensingalgorithmthatcancharacterizetherainfallduringthedayandnight.Inaddition,thispaperintroducesanintelligentwipercontrolalgorithmbasedonfuzzylogicthatcanregulatethewiperspeedandinterval.Finally,experimentalresultsunderthelaboratorysettingsarepresentedtoevaluatetheefficacyoftheproposedsystem.
Thispaperconsistsofsixsectionsincludingthisintroduction.SectionIIintroducesthebasicstructureofavision-basedsmartwipersystem,andSectionIIIpresentstherainsensingalgorithm.TheintelligentwipercontrolsystemusingfuzzylogicisdescribedinSectionIV,andtheexperimentalresultsaregiveninSectionV.Finally,theconclusionsarepresentedinSectionVI.
II.STRUCTUREOFVISION-BASEDSMARTWIPERSYSTEM
Inmanycases,thesensingareaofanopticalsmartwipersystemisverysmall;theareaisatmostI.OmmwideandIII0mmlong.Thismakesthewholesystemoperateonverylimitedinformation.Whenitstartstorain,thewiperbeginstomoveonlyaftersomeraindropsfallontothesensingarea.Moreover,thewipermaynot
startatallwhenwaterissplashedbyothercarsontotheareaotherthanthesensingarea.
Inordertoavoidtheseproblems,wedevelopedasmartwipersystemusingavisionsensorasshowninFig.I..Thesmartwipersystemconsistsofawipermotor,awiperswitch,afuzzycontroller,avisionsensor,andlightemittingdiodes(LEDs).Here,thewiperswitchisusedforactivatingthesmartwipersystem.Thefuzzycontrollerperformstwofunctions:therainsensingalgorithmthatcalculatestherainintensityandraindistributionfromtheimagecapturedbythevisionsensor,andthefuzzywipercontrolalgorithmthatcomputesthewiperspeedandwiperintervalbasedonrainintensityanddistribution.Finally,thevisionsensorisinstalledontheceilingofthevehicletocapturethewindshieldimage,andtheLEDsareinstalledinordertoprovidesomelightingduringthenighttime.
Toguaranteetheperformanceofthevision-basedsmartwipersystem,theexternalbackgroundoutsideofthevehiclemustbeignored,andonlytheraindropsonthewindshieldcanbeextractedfromtheimagecapturedbythevisionsensor.Forthispurpose,wehavefocusedthevisionsensoronthewindshieldwithlowdepthoffieldbyopeningtheaperturetothemaximum.Finally,thevisionsensorcanbeaffectedbyanexternallightsourcesuchassunlightinthedaytime,streetlamps,taillamps,andheadlampsinthenighttime.PartofdisturbancessuchassunlightandstreetlampscanbeavoidedbyinstallingthevisionsensortilteddownwardasshowninFig.I..However,otherdisturbancessuchastaillampsandheadlampscannotbeeliminated.Therefore,anappropriateimageprocessingalgorithmisnecessarytoeliminatetheeffectoftheseexternallightsources.Forthispurpose,wehaveanadditionalimageprocessingstepforthenighttime.
III.RAINSENSINGALGORITHMOFSMARTWIPERSYSTEM
Fig.IIshowstherainsensingalgorithmforthevision-basedsmartwipersystem.Thealgorithmtakesacapturedimageasitsinputandproducesthestateofprecipitationdescribedbytwovariables,i.e.,rainintensityandraindistributionasitsoutputs.Inordertoobtaintheoutputs,thealgorithmemploysaseriesofimageprocessingtechniques.Inthefigure,afterthewiperswitchissetto"auto"mode,thevisionsensoracquiresanimageofthewindshield.Then,dependingonwhethertheheadlampsareonoroff,thealgorithmdecideswhethertheadditionalnighttimeprocessingisrequired.Afterappropriateimageprocessingtechniquesareapplied,thealgorithmcomputesrainintensityanddistributionthatareusedtodeterminethewiperspeedandinterval.Therainintensityrepresentstheamountofrainonthewindshieldwhiletheraindistributionindicateshowwidelytheraindropsaredistributedonthewindshield.
III.I.Daytimeimageprocessingalgorithm
Inordertodetectraindropsonthewindshield,thealgorithmfindstheboundariesofraindropsbyusinganedgedetectiontechnique.AmongmanytechniquessuchasSobel,Prewitt,Roberts,Laplacian,andLaplacianofGaussian[V],theSobelmaskisselectedbecauseitsedgedetectioncapabilityisneithertoobadtodetectraindropsnortoogoodsothattheedgeofbackgroundisdetected.Fig.III(a)showsanoriginalimageofraindropswhileFig.III(b)showstheimageaftertheSobelmaskisapplied.Inthefigure,wecanseethattheboundariesofraindropsareclearwithoutanynoticeableedgesfromthebackground.
TheimageinFig.III(b)stillhastheboundariesofthebackground,thoughnotveryvisible,thatarerepresentedgraypixels.Thisisbecausethebackgroundimageisblurredduetothelowdepthoffield.Inordertoremovethesebackgroundboundariesandshortentheprocessingtime,asinglethresholdTisappliedasshowninEq.(I.)whereandaretheoldandnewvaluesforthepixel,respectively[VI].Inthisequation,thevalueforTisselectedasI.IIVaftersometrialsanderrors.
Fig.III(c)showstheimageafterthresholdingwherewecanseethatthegraybackgroundboundariesarecompletelygone.
Thenextstepistowhitenthepixelsinsidetheboundary.Forthispurpose,weusethedilationoperation[V]asanapproximation.Ingeneral,thedilationmakesanobjectenlargedbecausetheoperationextendstheoutermostpixelsoftheboundary.Therefore,theraindropswillbeslightlylargerandsomeofraindropsstillmayhaveblackpixelsinsidetheboundary.Inthiswork,aIIIXIIIdilationmask,whosepixelvaluesareallzeros,isusedforfastercomputationalspeed[VII].Fig.III(d)showsthefinalimageafterthedilationoperation.ComparingFig.III(c)withFig.III(d),wecanverifythatmostraindropboundariesarefilledwithwhitepixels.
III.IINighttimeimageprocessingalgorithm
Asmentionedearlier,thenighttimeimagestendtocontainmoredisturbancessuchasstreetlamps,taillamps,andheadlamps.Whenthedaytimeprocessingalgorithmisappliedtothenighttimeimage,theeffectofdisturbancesisstillremaininginthefinalimage.Thisindicatesthatthenighttimeprocessingalgorithmshouldbedifferentfromthedaytimealgorithm.
Forthispurpose,wefocusonthefactthatsharpfeatures,ingeneral,haverelativelyhighfrequencyelements,whileblurredfeatureshaverelativelylowfrequencyelements.Thatis,theraindropsundertheinternalLEDlightaredistinctandhavehighfrequencyelementswhereastheexternallightsourcesareblurredbythelowdepthoffieldandhavelowfrequencyelements.Therefore,disturbancesbytheexternallightsourcecanberemovedbyeliminatinglowfrequencyelementsfromtheimage.
Forremovinglowfrequencyelements,wechoosetheButterworthhighpassfilterthatcanbeexpressedasfollows:
WhereD(u,v)isthedistancefromtheoriginto(u,v)onthefrequencyplane.Inthiswork,thecutofffrequencyisselectedtobeIIIII,andnisselectedtobeI.VIviasometrialsanderrors.Fig.IV(a)showsanighttimeimagecontainingataillamp.Fig.IV(b)showstheimagewherethetaillampiseffectivelyremovedafterthehighpassfilterisappliedtotheoriginalimage.
III.IIICharacterizationofprediction
Thewiperoperationalgorithmrequiressomeinputvaluesderivedfromtheprocessedimage.Therefore,weneedastep,i.e.,characterizationofprecipitation,torepresenttheprocessedimagebysomevalues.Inordertoderivethesevalueswerelyonthenotionthatthefactorsthathinderthedriversviewaretheamountofwaterandthenumberofraindropsonthewindshield.First,whenthewindshieldiscoveredbythesignificantamountofwater,thedriverwillhavedifficultyinfindinglanesandobstacles.However,theamountof
waterisnottheonlyfactorbecausethedriversviewisconsiderablyblockedwhennumerousbutverysmallraindropsarecoveringthewindshield.
Therefore,wedefinetwovariables:rainintensityandraindistribution.Therainintensityisdefinedasthenumberofpixelsrepresentingraindrops(itsvalueis
high)intheprocessedimage(IIVVIxIIVVIpixel).Theraindistributioniscomputedasfollows:I.)dividetheoverallareaofIIVVIxIIVVIpixelimageintoI.VIxI.VIpixelunitareas,II)ifthereisatleastonepixelrepresentingaraindropintheI.VIxI.VIpixelunitarea,assignhighvaluetoallpixelsoftheI.VIxI.VIunitarea,III)theraindistributionisdefinedastheratioofthenumberofpixelswithhighvaluetothenumberofoverallpixels(VIV,VIIIVI).
Fig.Vshowsanexamplehowtherainintensityandraindistributionarecalculated.ThelargesquarerepresentstheIIVVIxIIVVIcapturedimagewhilethesmallsquarerepresentstheenlargedI.VIxI.VIunitarea.Fig.V(a)representsthecasefordrizzleormist.SupposethatraindropsareofthesizeofonepixelandtheyareevenlydistributedsothateachI.VIXI.VIunitareacontainsfourraindropsasshowninthesmallsquareinFig.V(a).Then,therainintensityisthenumberofraindroppixelswhichequalsI.0IIIV(IVineachunitareaI.VIxI.VIunitareasintheprocessedimage).TheraindistributionisI.00%becauseallpixelsofaunitareashouldhavehighvalueduetofourpixelsintheunitarea,andallunitareashavehighvalueduetotheevendistribution.Incontrast,Fig.V(b)representsthecaseofasinglebigraindropatthecenteroftheimage.Inthiscase,theraindropisaslargeasfourunitareasthatcontainI.0IIIVpixels.Therefore,therainintensityisI.0IIIV.TheraindistributionisI..VVI%becauseonlyfourunitareashavehighvalueoutofIIVVIunitareas(IVIIIVVIxI.00).
IV.FUZZYWIPERCONTROLALGORITHM
Inthispaper,afuzzycontrolalgorithmisusedtoregulatethewiperspeedandwiperintervalusingthecalculatedrainintensityanddistributionaftertheimageprocessing.Themajorrationaletousefuzzylogicisthatthefuzzylogicallowsmimickingthehumandriversdecisionmakingwithrelativeeaseandthattheoutcomesofthelogiccanbetunedorcustomizedforindividualdriversbyadjustingmembershipfunctions[VIII][IX].
Fig.VIshowsthestructureofthefuzzywipercontrolalgorithmofthesmartwipersystem.Inthefigure,thefuzzycontrollerconsistsofthreeparts:fuzzifier,inferenceengine,anddefuzzifier.Thefuzzifierconvertstherainintensityanddistributionintolinguisticvalues.Theinferenceenginecreatesthefuzzyoutputsusingfuzzycontrolrulesgeneratedfromexpertexperiences.Finally,thedefuzzifiercalculatesthewiperspeedandwiperintervalfromtheinferredresults.
Fig.VIIshowsthemembershipfunctionsofthefuzzyinputandoutputlinguisticvariables.ThelinguisticvariablesfortherainintensityaredefinedasDry,Drizzle,andRainwhilethosefortheraindistributionaredefinedasNarrow,Medium,andWideasshowninFig.VII(a).Inthefigure,themembershipfunctionsofrainintensityaredifferentdependingonwhetheritisdaytimeornighttime.Thisistocompensatetheboundarydetectioncapabilityinthenighttimebecausealmostalltheboundariesofraindropsaredetectedinthedaytimewhilethealgorithmtendstomisssomeraindropsinthenighttime.ThelinguisticvariablesforthewiperintervalaredefinedasZero,Short,andLongasshowninFig.VII(b).Here,wedefinethatthewiperintervalvariesfromzerotoIII,000msec.ThelinguisticvariablesforthewiperspeedaredefinedasZero,Low,andHighasshowninFig.VII(b).
Forfasterexecutionofthefuzzylogiccontroller,theMamdanismin-maxinferencemethodisused.Fordefuzzification,theMoM(Mean-of-Maximum)methodisusedforthewiperspeed,andtheCoA(Center-of-Area)methodisusedforthewiperinterval[I.0][I.I.].ThereasonforusingtheMoMmethodistoobtaintheoutputthat
changesdiscretelyinsteadofvaryingcontinuously.Thisisbecausethewipermotorsoperateattwodifferentspeeds,i.e.,highandlowinsteadofcontinuouslyvaryingspeed.Inthispaper,wehavethewiperstopiftheoutputvalueissmallerthanone,runathighspeediftheoutputvalueislargerthanIX,andrunatlowspeedotherwise.TheoutputmapsforspeedandintervalareshowninFig.VIII.
V.PERFORMANCEEVALUATIONOFVISION-BASEDSMARTWIPERSYSTEM
Inordertoevaluatetheefficacyofthevision-basedsmartwiper,wehavesetupanexperimentaltestbedinalaboratoryenvironment.Thetestbedhasawindshieldandawiperremovedfromacar.Theimageprocessingandfuzzylogicalgorithmsare
implementedonaPCwithaframegrabber(EuresysPicoloboard)andaCCDcamera(PulnixsTM-II00).ThePCisalsoconnectedtoaninterfacecircuitthatcandrivethewipermotor.Thecamerasapertureisadjustedtoopenaswideaspossible(f=I..IV).
V.I.Experimentalresultsfordaytimeoperation
Inordertoemulatethedrizzleinthedaytime,waterissprayedontothewindshieldwiththelaboratorylightsturnedon.Fig.IXshowstheimagesthroughouttheprocessing,i.e.,Sobelmask,thresholding,anddilationasexplainedearlier.ThefinalimagecontainsII,I.VIIIIIIwhitepixels(rainintensityofII,I.VIIIIII)andhasIII0.V%ofI.VI*I.VIunitareaswithatleastoneraindrops(raindistributionofIII0.V%).Usingthesetwovalues,wecancalculatethatthewiperintervalisII,IVV0msec,andthatthewiperspeedislowfromthefuzzywipercontrolalgorithm.Fig.I.0(a)showsthecapturedimageofthewindshieldwhenalargeamountofwaterisdroppedtoemulatetheraininthedaytime.Fromthefinalimage,therainintensityisI.V,VIIIIIII.pixels,andtherain
distributionisVIIIIII.II%.ThisgivesusIIIV0msecforwiperintervalandhighforwiperspeed.Comparingtheseresults,wecanseethatthefuzzylogicbehavesasintendedalthoughthemaximumwiperintervalofthreesecondsmaybetoosmallforrealapplications.
VI.SUMMARYANDCONCLUSION
Inthispaper,weproposedavision-basedsmartwipersystemtoimprovethesafetyandconvenienceofdrivers.Thesystememploysaseriesofimageprocessingstepstoremovebackgroundobjectsandidentifyraindropsonthewindshield.Fromtheprocessedimage,rainintensityanddistributionarecalculatedastwoinputstothe
fuzzylogicalgorithmselectingwiperspeedandinterval.Inordertodemonstratetheefficacyoftheproposedsystem,thesystemwasevaluatedthroughaseriesofexperimentsunderlaboratorysettings.Theconclusionsderivedfromtheresearchareasfollows.
Itisdemonstratedthatthevision-basedwipersystemistechnicallyviablebycombiningappropriateimageprocessingsteps.Thisvision-basedapproachprovidesfarmoreinformationthattheexistingoptics-basedsystems.Theopticalsensorssensingareamaybetoosmalltodetectthesparselydistributedraindropsandthesensorcannotdetecttheraindropsdirectlyonthelineofsightofdriver.Incontrast,thevisionsensorcancoveranarealargeenoughtodetectsparselydistributedraindropsandraindropsonthedriverslineofsight.
Rainintensityanddistributionmaybesuitablevariablestocharacterizearainfall.Theamountofwaterusedtobetheonlyvariabletocharacterizearainfallinexistingsystems.However,wehaveaddedanothervariable,raindistributiontorepresenthowraindropsaredistributedonthewindshield.Byusingthesetovariablesandasetofsimplefuzzyrules,wewereabletodeterminethewiperspeedandintermittentinterval.
Inspiteoftheaboveconclusions,thevision-basedwipersystemisnotstilleconomicallyviableduetotheextracostforcamera,framecapturecircuitry,andcontroller.Wehopethattheadvantagesofvision-basedwipersystemwilloutweightheextracostandcomplexityinthenearfuturewhenintelligentvehiclesbegintoappear.Whentheproposedsystembeginstomakesenseeconomically,theproposedalgorithmsshouldbetestedinthefieldandthelogicsshouldbefine-tunedbyobservingdriversbehavior.
附件II:外文原文
DevelopmentofVisionbasedControlSmartWindwhieldWiperSystemforIntelligentVehicle
Jee-HunParkMan-HoKimHong-JunIm
Kyung-ChangLeeandSukLee
DepartmentofMechanicalEngineering,PusanNationalUniversity,Busan,Korea
Abstract:Windshieldwipersplayakeyroleinassuringthedriverssafetyduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwiperintervalbecausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.Thispaperpresentsthedevelopmentofvision-basedsmartwindshieldwipersystemthatcanautomaticallyadjustitsspeedandintermittentintervalaccordingtotheamountofwaterdropsonthewindshield.Thesystememploysvariousimageprocessingalgorithmstodetectwaterdropsandfuzzylogictodeterminethespeedandtheintervalofthewiper.
Keywords:SmartWiperSystem,IntelligentVehicle,RainSensing,WiperControl,VisionSensor,FuzzyControl
I..INTRODUCTION
Intelligentvehiclesofferthepotentialtoenhancesafetyandconveniencesignificantlyforbothdriversandpassengers.Asacomponentoftheintelligenttransportationsystem(ITS),theintelligentvehicleusesvariousintelligentsensingandcontrolalgorithmstoassessthevehiclesenvironmentandassistthedriverwithsafedriving.Thesealgorithmsincludethedriverassistancesystemthatpartiallycontrolsthevehicleforthedriversconvenience,andthecollisionwarningsystemthatprovidesemergencyinformationtoallowthedrivertoavoidacollision.Becauseoftheirabilitytoenhancethesafetyandconvenience,intelligentvehiclesbecomeacrucialresearchareafortheintelligenttransportationsystem[I.][II].
Amongthevehiclepartsthatinfluencethedriverssafetyandconvenience,awindshieldwiperisanimportantpartthatallowsadrivertocollectvisualinformationduringprecipitation.Thetraditionalwipersystems,however,requiresdriversconstantattentioninadjustingthewiperspeedandtheintermittentwipervisionsensorinterval
becausetheamountofprecipitationonthewindshieldconstantlyvariesaccordingtotimeandvehiclesspeed.Becausethemanualadjustmentofthewiperdistractsdriversattention,whichmaybeadirectcauseoftrafficaccidents,manycompanieshavedevelopedautomaticwipersystemsusingsomeopticalsensorswithvariouslevelsofsuccess.
Manysmartwipersystemstrytoregulatethewipersspeedandintermittentintervalautomaticallyaccordingtotheamountofrainorsnow[III][IV].Thekeyelementofthesesystemsisthesensortomeasuretheamountofwateronthewindshield.Inmostsystems,anopticalsensorisusedforthispurpose[III].Thistypeofsensorsusesthefactthattherefractionangleandtheamountofreflectionofthelightaredifferentwhenthewindshieldiswet.ManysensorsuseseveralLEDstoemitthelightintothewindshieldatasmallincidentangle.Thelighttravelsthroughtheglasswithseveralreflectionsattheglass-airborder.OppositetotheLEDs,thereareseveralphotosensorstodetectthearrivinglightafterthereflections.Whenthewindshieldhaswateronitssurface,therefractionangleandtheamountofreflectedlightarechanged,resultinginthedifferentphotosensoroutput.Thisdifferenceisusedtodeterminetheamountofwateronthewindshield.
Theopticalrainsensorshavesomedisadvantageseventhoughtheyarewidelyused.Oneofdisadvantagesisthesensitivitytoexternallight.Thatiswhymanysystemshaveadarkfilteronthewindshieldsurfacewherethesensorisattachedinordertoblocktheexternallight.However,manysystemsstillactivatethewiperwhenthecarcomesoutoftunnelsorundergroundparkinglot.Anothershortfall,maybeamajoroneisthatthesensingareaisarelativelysmallportionofwindshield.Thismakesthewipersystemoperateonlywithlimitedinformation.Thewipersystemmayfailtoactivatewhentherearesomeraindropsonthedriverslineofsight,butnotonthesensingarea.
Forsolvingtheseproblems,thispaperpresentsavision-basedsmartwipersystemthatcanmeasurearelativelywiderwindshieldareathantheconventionalsystem.Especially,thevisionsensorhasadvantagethatcanmeasurenotonlytheamountofrainbutalsothedistributionofraindrops.Thisinformationcanallowthesystemmoreintelligentlywithoutdisturbingdrivers.
Inthispaper,wepresenttheconceptandkeyalgorithmsforavision-basedsmartwipersystem.Morespecifically,thispapershowsaneffectiverainsensingalgorithmthatcancharacterizetherainfallduringthedayandnight.Inaddition,thispaperintroducesanintelligentwipercontrolalgorithmbasedonfuzzylogicthatcanregulatethewiperspeedandinterval.Finally,experimentalresultsunderthelaboratorysettingsarepresentedtoevaluatetheefficacyoftheproposedsystem.
Thispaperconsistsofsixsectionsincludingthisintroduction.SectionIIintroducesthebasicstructureofavision-basedsmartwipersystem,andSectionIIIpresentstherainsensingalgorithm.TheintelligentwipercontrolsystemusingfuzzylogicisdescribedinSectionIV,andtheexperimentalresultsaregiveninSectionV.Finally,theconclusionsarepresentedinSectionVI.
II.STRUCTUREOFVISION-BASEDSMARTWIPERSYSTEM
Inmanycases,thesensingareaofanopticalsmartwipersystemisverysmall;theareaisatmostI.OmmwideandIII0mmlong.Thismakesthewholesystemoperateonverylimitedinformation.Whenitstartstorain,thewiperbeginstomoveonlyaftersomeraindropsfallontothesensingarea.Moreover,thewipermaynot
startatallwhenwaterissplashedbyothercarsontotheareaotherthanthesensingarea.
Inordertoavoidtheseproblems,wedevelopedasmartwipersystemusingavisionsensorasshowninFig.I..Thesmartwipersystemconsistsofawipermotor,awiperswitch,afuzzycontroller,avisionsensor,andlightemittingdiodes(LEDs).Here,thewiperswitchisusedforactivatingthesmartwipersystem.Thefuzzycontrollerperformstwofunctions:therainsensingalgorithmthatcalculatestherainintensityandraindistributionfromtheimagecapturedbythevisionsensor,andthefuzzywipercontrolalgorithmthatcomputesthewiperspeedandwiperintervalbasedonrainintensityanddistribution.Finally,thevisionsensorisinstalledontheceilingofthevehicletocapturethewindshieldimage,andtheLEDsareinstalledinordertoprovidesomelightingduringthenighttime.
Toguaranteetheperformanceofthevision-basedsmartwipersystem,theexternalbackgroundoutsideofthevehiclemustbeignored,andonlytheraindropsonthewindshieldcanbeextractedfromtheimagecapturedbythevisionsensor.Forthispurpose,wehavefocusedthevisionsensoronthewindshieldwithlowdepthoffieldbyopeningtheaperturetothemaximum.Finally,thevisionsensorcanbeaffectedbyanexternallightsourcesuchassunlightinthedaytime,streetlamps,taillamps,andheadlampsinthenighttime.PartofdisturbancessuchassunlightandstreetlampscanbeavoidedbyinstallingthevisionsensortilteddownwardasshowninFig.I..However,otherdisturbancessuchastaillampsandheadlampscannotbeeliminated.Therefore,anappropriateimageprocessingalgorithmisnecessarytoeliminatetheeffectoftheseexternallightsources.Forthispurpose,wehaveanadditionalimageprocessingstepforthenighttime.
III.RAINSENSINGALGORITHMOFSMARTWIPERSYSTEM
Fig.IIshowstherainsensingalgorithmforthevision-basedsmartwipersystem.Thealgorithmtakesacapturedimageasitsinputandproducesthestateofprecipitationdescribedbytwovariables,i.e.,rainintensityandraindistributionasitsoutputs.Inordertoobtaintheoutputs,thealgorithmemploysaseriesofimageprocessingtechniques.Inthefigure,afterthewiperswitchissetto"auto"mode,thevisionsensoracquiresanimageofthewindshield.Then,dependingonwhethertheheadlampsareonoroff,thealgorithmdecideswhethertheadditionalnighttimeprocessingisrequired.Afterappropriateimageprocessingtechniquesareapplied,thealgorithmcomputesrainintensityanddistributionthatareusedtodeterminethewiperspeedandinterval.Therainintensityrepresentstheamountofrainonthewindshieldwhiletheraindistributionindicateshowwidelytheraindropsaredistributedonthewindshield.
III.I.Daytimeimageprocessingalgorithm
Inordertodetectraindropsonthewindshield,thealgorithmfindstheboundariesofraindropsbyusinganedgedetectiontechnique.AmongmanytechniquessuchasSobel,Prewitt,Roberts,Laplacian,andLaplacianofGaussian[V],theSobelmaskisselectedbecauseitsedgedetectioncapabilityisneithertoobadtodetectraindropsnortoogoodsothattheedgeofbackgroundisdetected.Fig.III(a)showsanoriginalimageofraindropswhileFig.III(b)showstheimageaftertheSobelmaskisapplied.Inthefigure,wecanseethattheboundariesofraindropsareclearwithoutanynoticeableedgesfromthebackground.
TheimageinFig.III(b)stillhastheboundariesofthebackground,thoughnotveryvisible,thatarerepresentedgraypixels.Thisisbecausethebackgroundimageisblurredduetothelowdepthoffield.Inordertoremovethesebackgroundboundariesandshortentheprocessingtime,asinglethresholdTisappliedasshowninEq.(I.)whereandaretheoldandnewvaluesforthepixel,respectively[VI].Inthisequation,thevalueforTisselectedasI.IIVaftersometrialsanderrors.
Fig.III(c)showstheimageafterthresholdingwherewecanseethatthegraybackgroundboundariesarecompletelygone.
Thenextstepistowhitenthepixelsinsidetheboundary.Forthispurpose,weusethedilationoperation[V]asanapproximation.Ingeneral,thedilationmakesanobjectenlargedbecausetheoperationextendstheoutermostpixelsoftheboundary.Therefore,theraindropswillbeslightlylargerandsomeofraindropsstillmayhaveblackpixelsinsidetheboundary.Inthiswork,aIIIXIIIdilationmask,whosepixelvaluesareallzeros,isusedforfastercomputationalspeed[VII].Fig.III(d)showsthefinalimageafterthedilationoperation.ComparingFig.III(c)withFig.III(d),wecanverifythatmostraindropboundariesarefilledwithwhitepixels.
III.IINighttimeimageprocessingalgorithm
Asmentionedearlier,thenighttimeimagestendtocontainmoredisturbancessuchasstreetlamps,taillamps,andheadlamps.Whenthedaytimeprocessingalgorithmisappliedtothenighttimeimage,theeffectofdisturbancesisstillremaininginthefinalimage.Thisindicatesthatthenighttimeprocessingalgorithmshouldbedifferentfromthedaytimealgorithm.
Forthispurpose,wefocusonthefactthatsharpfeatures,ingeneral,haverelativelyhighfrequencyelements,whileblurredfeatureshaverelativelylowfrequencyelements.Thatis,theraindropsundertheinternalLEDlightaredistinctandhavehighfrequencyelementswhereastheexternallightsourcesareblurredbythelowdepthoffieldandhavelowfrequencyelements.Therefore,disturbancesbytheexternallightsourcecanberemovedbyeliminatinglowfrequencyelementsfromtheimage.
Forremovinglowfrequencyelements,wechoosetheButterworthhighpassfilterthatcanbeexpressedasfollows:
WhereD(u,v)isthedistancefromtheoriginto(u,v)onthefrequencyplane.Inthiswork,thecutofffrequencyisselectedtobeIIIII,andnisselectedtobeI.VIviasometrialsanderrors.Fig.IV(a)showsanighttimeimagecontainingataillamp.Fig.IV(b)showstheimagewherethetaillampiseffectivelyremovedafterthehighpassfilterisappliedtotheoriginalimage.
III.IIICharacterizationofprediction
Thewiperoperationalgorithmrequiressomeinputvaluesderivedfromtheprocessedimage.Therefore,weneedastep,i.e.,characterizationofprecipitation,torepresenttheprocessedimagebysomevalues.Inordertoderivethesevalueswerelyonthenotionthatthefactorsthathinderthedriversviewaretheamountofwaterandthenumberofraindropsonthewindshield.First,whenthewindshieldiscoveredbythesignificantamountofwater,thedriverwillhavedifficultyinfindinglanesandobstacles.However,theamountof
waterisnottheonlyfactorbecausethedriversviewisconsiderablyblockedwhennumerousbutverysmallraindropsarecoveringthewindshield.
Therefore,wedefinetwovariables:rainintensityandraindistribution.Therainintensityisdefinedasthenumberofpixelsrepresentingraindrops(itsvalueis
high)intheprocessedimage(IIVVIxIIVVIpixel).Theraindistributioniscomputedasfollows:I.)dividetheoverallareaofIIVVIxIIVVIpixelimageintoI.VIxI.VIpixelunitareas,II)ifthereisatleastonepixelrepresentingaraindropintheI.VIxI.VIpixelunitarea,assignhighvaluetoallpixelsoftheI.VIxI.VIunitarea,III)theraindistributionisdefinedastheratioofthenumberofpixelswithhighvaluetothenumberofoverallpixels(VIV,VIIIVI).
Fig.Vshowsanexamplehowtherainintensityandraindistributionarecalculated.ThelargesquarerepresentstheIIVVIxIIVVIcapturedimagewhilethesmallsquarerepresentstheenlargedI.VIxI.VIunitarea.Fig.V(a)representsthecasefordrizzleormist.SupposethatraindropsareofthesizeofonepixelandtheyareevenlydistributedsothateachI.VIXI.VIunitareacontainsfourraindropsasshowninthesmallsquareinFig.V(a).Then,therainintensityisthenumberofraindroppixelswhichequalsI.0IIIV(IVineachunitareaI.VIxI.VIunitareasintheprocessedimage).TheraindistributionisI.00%becauseallpixelsofaunitareashouldhavehighvalueduetofourpixelsintheunitarea,andallunitareashavehighvalueduetotheevendistribution.Incontrast,Fig.V(b)representsthecaseofasinglebigraindropatthecenteroftheimage.Inthiscase,theraindropisaslargeasfourunitareasthatcontainI.0IIIVpixels.Therefore,therainintensityisI.0IIIV.TheraindistributionisI..VVI%becauseonlyfourunitareashavehighvalueoutofIIVVIunitareas(IVIIIVVIxI.00).
IV.FUZZYWIPERCONTROLALGORITHM
Inthispaper,afuzzycontrolalgorithmisusedtoregulatethewiperspeedandwiperintervalusingthecalculatedrainintensityanddistributionaftertheimageprocessing.Themajorrationaletousefuzzylogicisthatthefuzzylogicallowsmimickingthehumandriversdecisionmakingwithrelativeeaseandthattheoutcomesofthelogiccanbetunedorcustomizedforindividualdriversbyadjustingmembershipfunctions[VIII][IX].
Fig.VIshowsthestructureofthefuzzywipercontrolalgorithmofthesmartwipersystem.Inthefigure,thefuzzycontrollerconsistsofthreeparts:fuzzifier,inferenceengine,anddefuzzifier.Thefuzzifierconvertstherainintensityanddistributionintolinguisticvalues.Theinferenceenginecreatesthefuzzyoutputsusingfuzzycontrolrulesgeneratedfromexpertexperiences.Finally,thedefuzzifiercalculatesthewiperspeedandwiperintervalfromtheinferredresults.
Fig.VIIshowsthemembershipfunctionsofthefuzzyinputandoutputlinguisticvariables.ThelinguisticvariablesfortherainintensityaredefinedasDry,Drizzle,andRainwhilethosefortheraindistributionaredefinedasNarrow,Medium,andWideasshowninFig.VII(a).Inthefigure,themembershipfunctionsofrainintensityaredifferentdependingonwhetheritisdaytimeornighttime.Thisistocompensatetheboundarydetectioncapabilityinthenighttimebecausealmostalltheboundariesofraindropsaredetectedinthedaytimewhilethealgorithmtendstomisssomeraindropsinthenighttime.ThelinguisticvariablesforthewiperintervalaredefinedasZero,Short,andLongasshowninFig.VII(b).Here,wedefinethatthewiperintervalvariesfromzerotoIII,000msec.ThelinguisticvariablesforthewiperspeedaredefinedasZero,Low,andHighasshowninFig.VII(b).
Forfasterexecutionofthefuzzylogiccontroller,theMamdanismin-maxinferencemethodisused.Fordefuzzification,theMoM(Mean-of-Maximum)methodisusedforthewiperspeed,andtheCoA(Center-of-Area)methodisusedforthewiperinterval[I.0][I.I.].ThereasonforusingtheMoMmethodistoobtaintheoutputthat
changesdiscretelyinsteadofvaryingcontinuously.Thisisbecausethewipermotorsoperateattwodifferentspeeds,i.e.,highandlowinsteadofcontinuouslyvaryingspeed.Inthispaper,wehavethewiperstopiftheoutputvalueissmallerthanone,runathighspeediftheoutputvalueislargerthanIX,andrunatlowspeedotherwise.TheoutputmapsforspeedandintervalareshowninFig.VIII.
V.PERFORMANCEEVALUATIONOFVISION-BASEDSMARTWIPERSYSTEM
Inordertoevaluatetheefficacyofthevision-basedsmartwiper,wehavesetupanexperimentaltestbedinalaboratoryenvironment.Thetestbedhasawindshieldandawiperremovedfromacar.Theimageprocessingandfuzzylogicalgorithmsare
implementedonaPCwithaframegrabber(EuresysPicoloboard)andaCCDcamera(PulnixsTM-II00).ThePCisalsoconnectedtoaninterfacecircuitthatcandrivethewipermotor.Thecamerasapertureisadjustedtoopenaswideaspossible(f=I..IV).
V.I.Experimentalresultsfordaytimeoperation
Inordertoemulatethedrizzleinthedaytime,waterissprayedontothewindshieldwiththelaboratorylightsturnedon.Fig.IXshowstheimagesthroughouttheprocessing,i.e.,Sobelmask,thresholding,anddilationasexplainedearlier.ThefinalimagecontainsII,I.VIIIIIIwhitepixels(rainintensityofII,I.VIIIIII)andhasIII0.V%ofI.VI*I.VIunitareaswithatleastoneraindrops(raindistributionofIII0.V%).Usingthesetwovalues,wecancalculatethatthewiperintervalisII,IVV0msec,andthatthewiperspeedislowfromthefuzzywipercontrolalgorithm.Fig.I.0(a)showsthecapturedimageofthewindshieldwhenalargeamountofwaterisdroppedtoemulatetheraininthedaytime.Fromthefinalimage,therainintensityisI.V,VIIIIIII.pixels,andtherain
distributionisVIIIIII.II%.ThisgivesusIIIV0msecforwiperintervalandhighforwiperspeed.Comparingtheseresults,wecanseethatthefuzzylogicbehavesasintendedalthoughthemaximumwiperintervalofthreesecondsmaybetoosmallforrealapplications.
VI.SUMMARYANDCONCLUSION
Inthispaper,weproposedavision-basedsmartwipersystemtoimprovethesafetyandconvenienceofdrivers.Thesystememploysaseriesofimageprocessingstepstoremovebackgroundobjectsandidentifyraindropsonthewindshield.Fromtheprocessedimage,rainintensityanddistributionarecalculatedastwoinputstothe
fuzzylogicalgorithmselectingwiperspeedandinterval.Inordertodemonstratetheefficacyoftheproposedsystem,thesystemwasevaluatedthroughaseriesofexperimentsunderlaboratorysettings.Theconclusionsderivedfromtheresearchareasfollows.
Itisdemonstratedthatthevision-basedwipersystemistechnicallyviablebycombiningappropriateimageprocessingsteps.Thisvision-basedapproachprovidesfarmoreinformationthattheexistingoptics-basedsystems.Theopticalsensorssensingareamaybetoosmalltodetectthesparselydistributedraindropsandthesensorcannotdetecttheraindropsdirectlyonthelineofsightofdriver.Incontrast,thevisionsensorcancoveranarealargeenoughtodetectsparselydistributedraindropsandraindropsonthedriverslineofsight.
Rainintensityanddistributionmaybesuitablevariablestocharacterizearainfall.Theamountofwaterusedtobetheonlyvariabletocharacterizearainfallinexistingsystems.However,wehaveaddedanothervariable,raindistributiontorepresenthowraindropsaredistributedonthewindshield.Byusingthesetovariablesandasetofsimplefuzzyrules,wewereabletodeterminethewiperspeedandintermittentinterval.
Inspiteoftheaboveconclusions,thevision-basedwipersystemisnotstilleconomicallyviableduetotheextracostforcamera,framecapturecircuitry,andcontroller.Wehopethattheadvantagesofvision-basedwipersystemwilloutweightheextracostandcomplexityinthenearfuturewhenintelligentvehiclesbegintoappear.Whentheproposedsystembeginstomakesenseeconomically,theproposedalgorithmsshouldbetestedinthefieldandthelogicsshouldbefine-tunedbyobservingdriversbehavior.
版权保护: 本文由 hbsrm.com编辑，转载请保留链接: www.hbsrm.com/lwqt/wxzs/104.html
