外文文献及译文Word文档下载推荐.docx
《外文文献及译文Word文档下载推荐.docx》由会员分享,可在线阅读,更多相关《外文文献及译文Word文档下载推荐.docx(24页珍藏版)》请在冰点文库上搜索。
FoodHandlingUsingComputer
文献、资料来源:
文献、资料发表(出版)日期:
2008.12
院(部):
机电工程学院
专业:
机械工程及自动化
班级:
机械054
******
学号:
**********
指导教师:
董明晓教授
翻译日期:
2009.5.15
外文文献:
FoodHandlingandPackagingusingComputervisionandRobot
Abstract
Eventhoughtheuseofrobotvisionsysteminmanufacturingsectorsisnowacommonplace,however,thetechnologyembodiedinthesedevicesispoorlymatchedtoindustrialneedsoffoodprocessors.Inparticular,foodprocessingimposesspecialdemandsuponmachinery.Forinstancethevisionsensormustbeprogrammedtodetectthepositionofsingleandisolatedobjectaswellasoverlappingoroccludingobjects.Specialgrippershavetobedesignedforhandlingoffoodarticlessuchthattheyhaveminimumcontactwiththefooditemsandhencecausingminimumdamagetothem.Inthisproject,startedoverayearago,avisionguidancesystemisbeingdevelopedtomeetthisobjective.ThesystemintegratesthemodifiedversionoftheHoughtransformalgorithmasthemainrecognitionengine.Themethodsandproceduresweretestedoncommerciallyproducedbeefburgers.
1.Introduction
Fromtheincomingdowntothepackaginglines,locating,recognizingandhandlingfoodobjectsareveryimportantinfoodprocessingindustry.Thesetasksareperformedroutinelyinfoodindustrymainlyforqualityevaluationandproductclassification.Suchtasksareverylaboriouslydemandingandtendtorelyheavilyonroleofthehumanoperator[1].Handsofworkersusingrawmaterialsofanimalorigincanheavilybecontaminatedwithfaecalandothermicro-pathogenicorganisms[2].ThestudybyTrickett[3]hasshownastronglinkbetweenfoodpoisoningandthehygienestandardsoffoodprocessors.Completeautomationoffoodhandlingandpackagingbymeansofroboticarmisthemosteffectivemeanstoeliminateinfluenceofmanualhandlingofmicrobiologicalqualityoffoods.
Robotshavesuccessfullybeenappliedinawiderangeoffoodindustriesprimarilydealingwithwell-definedprocessesandproductsnotonlybecausetheyarerelativelycleanandhygienic,alsobecauseoftheirflexibility,ruggednessandrepeatability.ThistrendwillcontinuetogrowwiththeincreasingscrutinyandregulatoryenforcementssuchasandHazardAnalysisandCriticalControlPoints(HACCP)togetherwithcompaniesthatarelookingforwaystodecreaseoreliminateworkerexposuretorepetitivemotiontasksandharshenvironment.Howeverthereareproblemsandchallengesassociatedwiththeuseofrobotsinfoodindustry[4].
Firstlythefoodproducts,despiteofthesametype,differinsize,shapeandotherphysicalvariables.Thisimposesspecialdemandsformachinerytohandlethem,requiringmultiplesensory,manipulationandenvironmentalcapabilitiesbeyondthoseavailableinrobotsdesignedtoautomatemanufacturingtasks.Secondlythesuccessofapplyingrobotsforfoodhandlers,hingesuponthesuccessofdetecting,locating,recognizingandhandlingseverelyoverlappingandoccludingcasesofsimilarfoodobjects.Thirdly,foodobjectsareoftendelicateandusuallycoveredwitheitherslipperyorviscoussubstances,makinghighspeedhandlingofsuchtargetsaverychallengingtask.Theexistingcontact-basedmechanismssuchasthevacuumsuctioningandtheclampgrippingarenotapplicablebecausetheycanpotentiallycauseinjuriesandbruisingtofoodproducts.Hencefurtherresearchisneededinordertosolvetheseproblems.Thispaperaddressessomeoftheproblems,focusingonthemethodsusedtocontroltherobotdirectlyfromthevisionsensor,attemptingtosimulatethewaythathumansusetheireyestonaturallycontrolthemotionoftheirarms.
2.MaterialsandMethods
2.1SamplePreparation
Thechosenfoodforthisstudyisalocallyproducedbeef-burger.Itpossessesalltheimportantcharacteristicswhichareuniquetofoodproducts,suchthattheyareveryfragileandeasilydeformed.Theaveragesizeofthebeef-burgersis8.5mminthicknessand46.1mmradiusand69.3gminweight.Surfaceimagesoftestsampleswereacquiredusing8-bitrobotvisionsystemwithuniformwhitebackground.Thewhitebackgroundprovidesexcellentcontrastbetweentheburgerandthebackground.Thechosenexposurewasadjustedsothattheimageintensityhistogramswereapproximatelycenteredatmid-wayofthefull-scalerange.Thefocaldistancewasselectedtoallowsingleaswellasmultiplesamplestofitintheimageframe.
2.2Robotvision
TherobotvisionsystemsusedinthisstudyistheAdeptCobra6004-DOFarticulatedscararobot,manufacturedbyAdeptTech.,USAandequippedwithAdeptVisionInterface,MV-5AdeptcontrollerandTM1001CCDmonochromecameramanufacturedbyPulnixInc.,Canada.Thecamerawasmountedontolink2ofrobotarmandilluminatedusingthewarmwhitedeluxe(WWX)fluorescentlighting.ThecameraisfittedwithaC-mountadaptertopermittheuseofTamronf/25.58-mmlens.TheTM1001cameraisconnectedtotheAVIcardviaa12-pinHirosetypecameraconnectorofHiroseInc.,Japan.TherobotvisionsystemwasoperatedusingAdept’sAIMSv4.0andprogramminglibraries,runningon1.7GHzand255MBRAMPentiumIVPC.Figure1showstheset-upofrobotvisionsystem.
2.3ImageProcessing
Theobjectiveofimageprocessinginrobotvisionapplicationsismainlytoextractmeaningfulandaccurateinformationfromtheimages,endowingtherobotswithmoresophisticatedpositioncontrolcapabilitiesthroughtheuseofvisionfeedback.Theuseofasimplegeometricmethodsuchasintroducingspeciallydesignedcuesintotheimagescenewillnotworkinthisapplicationsincetheburgerimagesaregenerallycomplex,difficulttomodelandpartiallyorextensivelyoccludeddependingontheviewingangle.Figure2showsthetypicalbeef-burgerimage.
Inordertoaccuratelytranslateburgerpositionstorobotmovements,theformergeometricfeaturesmustfirstlybeextractedandsecondlymatchedtotherobot'
sworkspace.Inthisapplicationoneoftheusefulfeatureswhichuniquelycharacterizetheposeofaburgerinarbitrarylocationsisitscentroid.Thisgeometricdescriptorisapplicablesincetheshapeofaburgerisapproximatelycircular.Furthermorethisfeaturepreservesvariancetotranslation,rotationandscaling.Beforecomputingthecentroidoftherealburgerimages,severalpreprocessingoperationsneedtobeperformedoneachimage.
Edgedetectionoperationiscarriedouttodetectthecontouroftheconnectedandisolatedcomponents,thereby,effectivelytransformingtheoriginaldataintoaformsuitableforfurtherprocessing.TheedgeresultsofFigure2computedusingwell-knownSobelandRobertoperators[5]areshowninFigures3(a),(b),(c)&
(d).Fromthesefiguresitcanbeseenthattheedgesdeterminedbytheseoperatorscomprisedofmanyfalseedges,discontinuitiesandspuriousspotsresultingfromunevenandirregularsurfaceoftheburger,non-uniformlightreflectionandshadows.Thesedrawbacksarenotacceptableforapplicationdescribedinthispaper.
Amoresophisticatedmethodisneededinordertoobtainacceptableresults.ThemethodusedtosolvetheseproblemswasbasedonCannyedgedetectionoperator[6].Interestedreadersarereferredtothispublicationfordetailedmathematicalexplanationofthisrelativelynewedgedetector.Hereonlytheimportantprinciplesarepresentedinordertofacilitatediscussiononrobotvisionapplicationsonfoodhandling.Cannymethodforedgedetectionisprincipallybasedonsomegeneralideas.
FirstlyCannywasthefirsttodemonstratethatconvolvinganimagewithasymmetric2-DGaussianfilterandthen,secondly,differentiatinginthedirectionofthegradientformtheedgemagnitudeimage.Thepresenceofedgesintheoriginalimagegivesrisetoridgesingradientmagnitudeimage.Theobjectiveistodetectthetrueedgeintherightplace.Thiscanbedoneusingmethodknownasnon-maximalsuppressiontechnique.Essentiallythismethodworksbytrackingalongthetopoftheridges,retainingonlythosepointsatthetopoftheridge,whilstsuppressingallothers.Thetrackingprocessexhibitshysteresiscontrolledbytwoimportantparameters.TheyarethelowerthresholdvalueTlow,andtheupperthresholdvalueThigh.IftheedgeresponseisaboveThigh,thenthispixeldefinitelyconstitutesanedgeandhenceretained.PixelslessthanThighbutgreaterthangreaterTlowareconsideredasweakedges.Finallytrackingwasdonetobridgealldiscontinuededgesaswellastoeliminatethefalseones.Weakedgesareretainedonlyiftheyareconnectedtothestrongedge.Theresultoftheseoperationsisanimagewiththinlinesofedgepointswithimprovededge-to-noiseratio.Eventhoughthismethodreducestheeffectofnoise,however,theoverallqualityofedgesdependslargelyontheoptimalselectionofthestandarddeviationı.whichdefinestheGaussianmaskforCanny’sedgedetection.Experimentallytheoptimumvaluewassetto3.Thiscorrespondstoa25X25kernel.Thisvalueisfixedforgivensetofbackgroundilluminationandimagegain.Changeinanyoftheseexternalfactorssuchasillumination,imagegain,backgroundcolourwillalsoaffecttheoptimumvalueofı.Figures4(a)&
(b)showresultsforcannyedgedetectionwithısetto1and3.
ComparingFigure3andFigure4,itcanbeseenclearlythattheedgesdeterminedbyCanny'
soperatorarelesscorruptedcomparedtoedgesdetectedeitherbySobelorRobertoperator.TheburgeredgesaremorecompleteinFigure4whereasinFigure3theyareonlypartiallyvisibleandmoreobscured.FurthermoretheretentionofmajordetailbytheCannyoperatorisveryevident.Thepresenceofoverlappingandpartiallyocclud