Advances in Machine Learning and Data Analysis Lecture Notes inElectrical Engineering Volume48 Forothertitlespublishedinthisseries,goto http://www.springer.com/series/7818 Sio-Iong Ao (cid:129) Burghard B. Rieger Mahyar Amouzegar Editors Advances in Machine Learning and Data Analysis 123 Editors Sio-IongAo MahyarAmouzegar HarvardSchoolofEngineering CollegeofEngineering andAppliedSciences CaliforniaStateUniversity HarvardUniversity LongBeach Room403,60OxfordStreet CambridgeMA02138,USA [email protected] BurghardB.Rieger UniversitätTrier FBIILinguistische Datenverarbeitung Computerlinguistik Universitätsring15 54286Trier Germany [email protected] ISSN1876-1100 e-ISSN1876-1119 ISBN978-90-481-3176-1 e-ISBN978-90-481-3177-8 DOI10.1007/978-90-481-3177-8 SpringerDordrechtHeidelbergLondonNewYork LibraryofCongressControlNumber:2009930421 (cid:2)c SpringerScience+BusinessMediaB.V.2010 Nopartofthisworkmaybereproduced,storedinaretrievalsystem,ortransmittedinanyformorby anymeans,electronic,mechanical,photocopying,microfilming,recordingorotherwise,withoutwritten permissionfromthePublisher,withtheexceptionofanymaterialsuppliedspecificallyforthepurpose ofbeingenteredandexecutedonacomputersystem,forexclusiveusebythepurchaserofthework. Coverdesign:eStudioCalamarS.L. Printedonacid-freepaper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) Preface A large international conference on Advances in Machine Learning and Data Analysis was held in UC Berkeley, CA, USA, October 22–24, 2008, under the auspices of the World Congress on Engineering and Computer Science (WCECS 2008). The WCECS is organized by the International Association of Engineers (IAENG).IAENGisanon-profitinternationalassociationfortheengineersandthe computerscientists,whichwasfoundedin1968andhasbeenundergoingrapidex- pansionsinrecentyears.TheWCECSconferenceshaveservedasexcellentvenues for the engineering community to meet with each other and to exchange ideas. Moreover, WCECS continues to strike a balance between theoretical and appli- cation development.The conferencecommittees have been formed with over two hundredmemberswhoaremainlyresearchcenterheads,deans,departmentheads (chairs),professors,andresearchscientistsfromoverthirtycountries.Theconfer- enceparticipantsarealsotrulyinternationalwithahighlevelofrepresentationfrom manycountries.Theresponsesforthecongresshavebeenexcellent.In2008,were- ceivedmorethansixhundredmanuscripts,andafterathoroughpeerreviewprocess 56.71%ofthepaperswereaccepted. This volume contains sixteen revised and extended research articles written by prominent researchers participating in the conference. Topics covered include Expertsystem,Intelligentdecisionmaking,Knowledge-basedsystems,Knowledge extraction, Data analysis tools, Computational biology, Optimization algorithms, Experimentdesigns,Complexsystemidentification,Computationalmodeling,and industrialapplications.Thebookoffersthestateoftheartoftremendousadvancesin machinelearninganddataanalysisandalsoservesasanexcellentreferencetextfor researchersandgraduatestudents,workingonmachinelearninganddataanalysis. HarvardUniversity,USA Sio-IongAo UniversityofTrier,Germany BurghardB.Rieger CaliforniaStateUniversity,LongBeach,USA MahyarAmouzegar v Contents 1 2D/3DImageDataAnalysisforObjectTrackingand Classification.................................................................. 1 Seyed Eghbal Ghobadi, Omar Edmond Loepprich, Oliver Lottner, Klaus Hartmann, Wolfgang Weihs, andOtmarLoffeld 2 Robot Competence Development by Constructive Learning....................................................................... 15 Q.Meng,M.H.Lee,andC.J.Hinde 3 UsingDigitalWatermarkingforSecuringNextGeneration MediaBroadcasts............................................................. 27 DominikBirkandSea´nGaines 4 AReduced-DimensionProcessorModel................................... 43 AzamBeg 5 Hybrid Machine Learning Model for Continuous MicroarrayTimeSeries...................................................... 57 Sio-IongAo 6 An Asymptotic Methodto a FinancialOptimization Problem........................................................................ 79 DejunXie,DavidEdwards,andGibertoSchleiniger 7 AnalyticalDesignofRobust Multi-loopPIController forMulti-timeDelayProcesses.............................................. 95 TruongNguyenLuanVuandMoonyongLee 8 AutomaticandSemi-automaticMethodsfortheDetection ofQuasarsinSkySurveys...................................................109 Sio-IongAo vii viii Contents 9 ImprovingLow-CostSailSimulatorResultsbyArtificial NeuralNetworksModels ....................................................139 V.D´ıazCasa´s,P.PorcaBel´ıo,F.Lo´pezPen˜a,andR.J.Duro 10 RoughSetApproachestoUnsupervisedNeuralNetwork BasedPatternClassifier......................................................151 AshwinKothariandAvinashKeskar 11 A NewRobust CombinedMethodforAutoExposure andAutoWhite-Balance.....................................................165 QuocKienVuong,Se-HwanYun,andSukiKim 12 A Mathematical Analysis Around Capacitive Characteristicsof the Current of CSCT: Optimum UtilizationofCapacitorsofHarmonicFilters.............................179 MohammadGolkhahandMohammadTavakoliBina 13 Harmonic Analysis and Optimum Allocation ofFiltersinCSCT ............................................................191 MohammadGolkhahandMohammadTavakoliBina 14 Digital Pen and Paper Technology as a Means ofClassroomAdministrationRelief........................................203 JanBroer,TimWendisch,andNinaWillms 15 AConceptualModelforaNetwork-BasedAssessment SecuritySystem...............................................................217 NathanPercival,JenniferPercival,andClemensMartin 16 Incorrect Weighting of Absolute Performance inSelf-Assessment............................................................231 ScottA.JeffreyandBrianCozzarin Chapter 1 2D/3D Image Data Analysis for Object Tracking and Classification SeyedEghbalGhobadi,OmarEdmondLoepprich,OliverLottner, KlausHartmann,WolfgangWeihs,andOtmarLoffeld Abstract Object tracking and classification is of utmost importance for different kindsofapplicationsincomputervision.Inthischapter,weanalyze2D/3Dimage data to addresssolutionsto someaspects ofobjecttrackingandclassification. We concludeourworkwitharealtimehandbasedrobotcontrolwithpromisingresults inarealtimeapplication,evenunderchallengingvaryinglightingconditions. Keywords 2D/3Dimagedata(cid:2)Registration(cid:2)Fusion(cid:2)Featureextraction(cid:2)Tracking (cid:2)Classification(cid:2)Hand-basedrobotcontrol 1.1 Introduction Object tracking and classification are the main tasks in different kinds of appli- cations such as safety, surveillance, man–machine interaction, driving assistance systemandtrafficmonitoring.Ineachoftheseapplications,theaimistodetectand findthepositionofthedesiredobjectateachpointintime.Whileinthesafetyap- plication, the personnel as the desired objects should be tracked in the hazardous environmentstokeepthemsafefromthemachinery,inthesurveillanceapplication theyaretrackedtoanalyzetheirmotionbehaviorforconformitytoadesirednorm forsocialcontrolandsecurity.Man-Machine-Interaction,ontheotherhandhasbe- comeanimportanttopicfortheroboticcommunity.Apowerfulintuitiveinteraction betweenmanandmachinerequirestherobottodetectthepresenceoftheuserand interprethisgesturemotion.Adrivingassistancesystemdetectsandtrackstheob- stacles,vehiclesandpedestriansinordertoavoidanycollisioninthemovingpath. Thegoaloftrafficmonitoringinanintelligenttransportationsystemistoimprove theefficiencyandreliabilityofthetransportsystemtomakeitsafeandconvenient S.E.Ghobadi((cid:2)),O.E.Loepprich,O.Lottner,K.Hartmann,W.Weihs,andO.Loffeld CenterforSensorSystems(ZESS),UniversityofSiegen,Paul-Bonatz-Str.9-11, D57068Siegen,Germany e-mail:[email protected];[email protected];[email protected]; [email protected];[email protected];[email protected] S.-I.Aoetal.(eds.),AdvancesinMachineLearningandDataAnalysis, 1 LectureNotesinElectricalEngineering48,DOI10.1007/978-90-481-3177-8 1, (cid:2)c SpringerScience+BusinessMediaB.V.2010 2 S.E.Ghobadietal. for the people. There are still so many significant applications in our daily life in whichobjecttrackingandclassificationplaysanimportantrole.Nowadays,the3D visionsystemsbasedonTimeofFlight(TOF)whichdeliverrangeinformationhave the main advantage to observe the objects three-dimensionallyand therefore they havebecomeveryattractivetobeusedintheaforementionedapplications.However, the current TOF sensors have low lateral resolution which makes them inefficient foraccurateprocessingtasksintherealworldproblems.Inthiswork,wefirstpro- poseasolutiontothisproblembyintroducingournovelmonocular2D/3Dcamera system and then we will study some aspects of object tracking and classification using2D/3Dimagedata. 1.2 2D/3DVisionSystem AlthoughthecurrentopticalTOFsensors [13–16]canprovideintensityimagesin additiontotherangedata,theysufferfromalowlateralresolution.Thisdrawback can be obviated by combininga TOF camera with a conventionalone. This com- binationisatendencyintherecentresearchworksbecauseevenwithregardtothe emerging new generation of TOF sensors with high resolution,1 an additional2D sensorstillresultsinahigherresolutionandprovidesadditionalcolorinformation. With regardto the measurementrange,however,the problemof parallaxdoesnot allowtosimplypositiontwocamerasnexttoeachotherandoverlaythegenerated images. The multimodal data acquisition device used in this work is a recently devel- oped monocular 2D/3D imaging system, named MultiCam. This camera, which is depicted in Fig.1.1, consists of two imaging sensors: A conventional 10-bit CMOSgrayscalesensorwithVGAresolutionandaPhotonicMixerDevice(PMD) with a resolution of 64(cid:3)48 pixels. The PMD is an implementationof an optical Time of Flight (TOF) sensor, able to deliver range data at quite high frame rates. VIS -NIR beam splitter F-M(oRuenatr C Aapmeretruar eL)ens VISA-NR@I RC t7Bd6.e 0faonmrm Ns,pIRlitter3D PMD Sensor forC-M(oRuenatr CAapmereturare L)ens AR@ C t7d6. 0fonrm N,IR3D PMD Sensor for Camera Body Lens Infrared Lighting modulated NIR modulated NIR glass Sensor Window glass Sensor Window AR Ctd. for NIR AR Ctd. for NIR NIR Edge Filter NIR Edge Filter @ 1000nm @ 1000nm 2D CMOS Sensor IR Cut @ 730nm 2D CMOS Sensor IR Cut @ 730nm for VIS (optional) for VIS (optional) Fig. 1.1 Left: 2D/3D vision system (MultiCam) developed at ZESS. Middle: F-mount optical setup.Right:C-mountopticalsetup 1ForexamplePMD-40K (200(cid:3)200pixels), Swissranger 4000(176(cid:3)144pixels) andZCam- prototype(320(cid:3)480pixels).