高可靠性车辆检测与车道碰撞预警系统(IJWMT-V8-N2-1)

I.J.WirelessandMicrowaveTechnologies,2018,2,1-14PublishedOnlineMarch2018inMECS()DOI:10.5815/ijwmt.2018.02.01Availableonlineat:19January2018;Accepted:13February2018;Published:08March2018AbstractInthelasttwodecades,developingDrivingAssistanceSystemsforsecurityhasbeenoneofthemostactiveresearchfieldsinordertominimizetrafficaccidents.Vehicledetectionisavitaloperationinmostoftheseapplications.Inthispaper,wepresentahighreliableandreal-timelighting-invariantlanecollisionwarningsystem.Weimplementanovelreal-timevehiclesdetectionusingHistogramofOrientedGradientandSupportVectorMachinewhichcouldbeusedforcollisionprediction.Thus,inordertomeettheconditionsofreal-timesystemsandtoreducethesearchingregion,Otsu’sthresholdmethodplayacriticalroletoextracttheRegionofInterestusingthegradientinformationfirstly.Secondly,weuseHistogramofOrientedGradient(HOG)descriptortogetthefeaturesvector,andthesefeaturesareclassifiedusingaSupportVectorMachine(SVM)classifiertogettrainingbase.Finally,weusethisbasetodetectthevehiclesintheroad.Twosetsgeneratedthetrainingdataofoursystemasetofnegativeimages(non-vehicles)asetofpositiveimages(vehicles),andthetestisperformedonvideosequencesontheroad.Theproposedmethodologyistestedindifferentconditions.Ourexperimentalresultsandaccuracyevaluationindicatestheefficiencyofyoursystemproposedforvehiclesdetection.IndexTerms:DrivingAssistanceSystems(DAS);FrontCollisionWarningSystem(FCWS);Otsuthreshold;HistogramofOrientedGradient(HOG);SupportVectorMachine(SVM).©2018PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience1.IntroductionDrivingAssistanceSystems(DAS)suchasfrontcollisionandlanedeparturewarninghasincreasessafeand*Correspondingauthor.YassinKortliE-mailaddress:[e-mail]2High-reliabilityVehicleDetectionandLaneCollisionWarningSystemsecuredriving.Thissystemusedtoadjust,enhance,andautomatethedriving.Themajorityoftrafficaccidentshappenbecauseofdriverslackattention.DrivingAssistanceSystemsreducesthedriverworkloadandprovidessecurity.Thesystemeitheralertsthedriverwheneveradangeroussituationisencountered.This,thispaperfocusonimplementingFrontCollisionWarningSystem(FCWS)inrealtime.TheFCWSisanimportantmoduleinresearchanddevelopmentofIntelligentTransportationSystems(ITS).FCWSbasedonmonocularvision,seeitselfasakeytoavoidingdeathsbyaccidentwithhighreliabilityandlowcost.Differentsystemshavebeenimplementedintheliteratureinordertodetectthevehiclesinfrontandenhancethesecurity.ontheroad.Allofthesetechniquescanbeevaluatedusingtwoseparatephases:thelearningphaseandthedetectionphase.Moreover,anFCWSpresentthreemajorprocessingmodules[1]:Preprocessing.Featureextractiontostorethefeaturesextractedfromdatasets(vehiclesandnovehicles).Detectionbymatchingthefeaturesstoredinthesystemdatasetswiththefeaturesextractedfromthevehiclestestinginthevideossequences.Manyalgorithmshavebeenimplementedinordertodetectfaces.ViolaandJones[7]implementedVIOLAJONESdescriptorisusedforfacedetection.Thisapproachisdonewithtowtechniques;theHaarfeaturesusedtoextractthefeaturesvectorofthefacesandAdaboostclassifierfordetectionfaces.But,thisapproachpresentshighcomputationalcomplexitiesanddoesnotsolvetheilluminationproblem.LocalBinaryPatterns,SURF,SIFT,andGaborfilter[3]aresomeofmanyothertechniquesquotedintheliterature.However,thebiggestproblemofthesetechniquesisunsatisfactoryperformanceandhighcomputationalcomplexitiesundervariouslightingconditions.Inthiswork,weproposetocombinetheuseofHistogramOfOrientedGradient(HOG)descriptor[9]withSVMclassifierforvehicledetection.HistogramOfOrientedGradient(HOG)descriptorisanadequatedescriptorintermsofrobustnessanddiscriminationtoextract“gradientdistribution”vehiclefeaturescomparedwiththeHaardescriptor[7].Asweexplainedbellow,thistechniqueisconsideredsincetheycanbeusedforlearninganddetectionprocesses.Inthisresearch,themajorcontributionsofourworkcanbesummarizedasfollows:WeappliedtheOTSUsegmentationapproachtosegmenttheinputframesandtocopewiththelightingproblem.Thismethodisrobustinthepresenceofilluminationconditions,shadow,noiseandlackoflanepainting.AccordingtothegradientinformationobtainedbytheOTSUsegmentationmethod,weproposetodetermineasimpleadaptiveROIbyusingahorizonline,aimingatreducingthecomputationalcomplexityduetotheprocessingtime.Processingentirepixelsofthefullimageisunnecessary.HistogramofGradient(HOG)combinedwithSVMclassifierisanadequatetechniqueintermsofdecreasingthefalsepositiverateaswellasreducingtheamountofcomputationnecessarytoextract“gradientdistribution”vehiclefeaturesandthesefeaturesarethenusedtotrainwithSupportVectorMachine(SVM)classifiertogetatrainingbase.Thetrainingbaseisusedtodetectthevehiclesintheroadand,ou