39 10 : PWM Fig.1 Flow ChartofCrowd MonitoringApproachBasedonPWM (pixel)d r (mm) 2(b) Q r A D r (1) : D r = H/ cosθ = H/ cos rε f (1) ε CCD Q r

Size: px

Start display at page:

Download "39 10 : PWM Fig.1 Flow ChartofCrowd MonitoringApproachBasedonPWM (pixel)d r (mm) 2(b) Q r A D r (1) : D r = H/ cosθ = H/ cos rε f (1) ε CCD Q r"

匠季
5 years ago
Views:

1 GeomaticsandInformationScienceofWuhanUniversity Vol.39No.10 Oct.2014 DOI: /j.whugis : (2014) : : ; ; P231.5 :A ROI : 1.2 [1-3] [4] [5-9] [9] :1 ;2 (perspective ;3 weightmodelpwm) 1 PWM PWM f H : A θ AQ RG (regionofinterestroi) [10] (a) G g G A Q r ROI qrq r 2(b) 2(b) P G P r P R qr qr r : : ( ); (Q ) : huxuemin2003@163.com

2 39 10 : PWM Fig.1 Flow ChartofCrowd MonitoringApproachBasedonPWM (pixel)d r (mm) 2(b) Q r A D r (1) : D r = H/ cosθ = H/ cos rε f (1) ε CCD Q r A 2(b) P G P G P G w G=1 P G r w r : w r = Dr = sec rε (2) D G f 2 (2) Fig.2 2D Modeland3D ModelofFisheyeCamera X = w r (xy) (3) (x y) F w r (xy) (xy) ;X (weightedtangentialgradientwtg) ;F (3) :

3 X (weightedarea WA) (weightedcontourscamera perimeterwcp) (weightededge pixelnumber WEPN) Harris (weightedharriscornernumberwhcn) F ROI [11] 3 Canny Fig.3 A MethodofComputingtheTangential Harris [12] GradientandtheRadialStructureElement WAGR =WA/WTG 3(a) P 6 (xy)p 1~P 8 P 8 O (x c yc) (4) : G t (P)= I(P 1 )+2I(P 2 )+I(P 3 ) (4) -I(P 5 )-2I(P 6 )-I(P 7 ) I ;G t ROI (b) ROI P P P 1 P 2 O (5) : 6 P 1 P 2 y P 1 P 2 n =^β 0 + ^βjx j (5) j=1 n ^β0~^β 6 [15] P 6 WA WCP WEPN WHCN WTG WAGR (5) x 1 : ~x 6 y 1) ; 2) ; X={WAWCPWEPNWHCNWTG WAGR} P 2 P 6 PO P O 1.4 P 1~P 8 [1314] Sobel P AdaBoost : 2 3) 1 n 4) (3) (weighted area-gradient ratio WAGR ) (6)

39 10 : 1239 4 N f V err = ( fu(e(i)e t (i)))/n f 100% (6) i=1 V err ;E(i) i ;E t (i) i ;N f ; fu E(i) E t (i) 0 1 (7) N f M err = ( Err(i))/N f = i=1 N f ( y(i)-yt(i) )/N f 100% (7) i=1 y(i) M err

4 39 10 : N f V err = ( fu(e(i)e t (i)))/n f 100% (6) i=1 V err ;E(i) i ;E t (i) i ;N f ; fu E(i) E t (i) 0 1 (7) N f M err = ( Err(i))/N f = i=1 N f ( y(i)-yt(i) )/N f 100% (7) i=1 y(i) M err ;Err(i) i ;y(i) i ( );yt(i) i (a)~ 4(h) 1 (1)~(8) PWM 8 1 PWM Tab.1 ResultsofCrowdDensityRatingandCounting (1) (2) (3) (4) (5) (6) (7) (8) /% /ms /% /% /ms Fig.4 ResultsofDensityEstimationandCrowdCountingoftheSixTestingSamples

5 [8] GuoSenYan HepingLiu Wei.Crowd Density EstimationBasedonGreyLevelDependenceMatrix [J].Computer Engineering and Applications (28): (. [J] (28): ) [9] Hu XueminZheng HongZhang Qing.Crowd MonitoringforUndergroundRailwayStationBased ( on Weighted AreaPerspectiveTransformation[J]. ) GeomaticsandInformation Scienceof Wuhan Uni- versity201237(3): (. [1] Paisitkriangkrai SShen ChunhuaZhang Jian. FastPedestrianDetectionUsingaCascadeofBoos- tedcovariancefeatures[j].ieee Transactionson Circuitsand Systemsfor Video Technology (8): [2] Sim C HRajmadhanERanganathS.A Two-Step Approachfor DetectingIndividuals Within Dense Crowds[C].The5thInternationalConferenceon Articulated MotionandDeformableObjectsSinga- pore2008 [3] WangZhenruiJiaYulanHuang Hua etal.pe- destrian Detection Using Boosted HOG Features [C].The11thInternationalIEEE Conferenceon Inteligent Transaction SystemsBeijing China 2008 tiontechniquewithanapplicationtostereovision tificialinteligencesanfranciscousa1981 [5] MaranaA NCostaLFLotufoR Aetal.Onthe EficacyofTextureAnalysisforCrowd Monitoring [4] LucasBKanadeT.AnIterativeImageRegistra- [C].The7thInternationalJointConferenceonAr- [C].InternationalSymposiumonComputerGraph- icsimageprocessingandvisionriodejaneiro Brazil1998 [6] ChanA BLiangZJVasconcelosN.PrivacyPre- servingcrowd Monitoring:CountingPeople With- outpeople ModelsorTracking[C].IEEE Confer- enceoncomputervisionandpaternrecognition Anchorage2008 [7] KongDanGrayDHaiTao.A ViewpointInvari- antapproachforcrowdcounting[c].the18thin- ternational Conference on Patern Recognition HongKongChina2006 [J] (3): ) [10]Liu KZhouF QZhang GJ.RadiusConstraint LeastSquareCircleFiting MethodandErrorAnal- ysis[j].journalof Optoelectronics & Laser (5): [11]GionnisP.AutomaticContourExtractionfrom2D Image[J].Applied MedicalInformatics (1):9-15 [12]LoweD G.DistinctiveImageFeaturefromScale-In- variant Keypoints[J].InternationalJournalof Computer Vision (2): [13]FreundYSchapireR E.ExperimentswithaNew Boosting Algorithm [C].InternationalConference on MachineLearningSanFranciscoUSA1996 tedclassificationofhighspatial-resolutionremote SensingImageryBasedonAdaBoost[J].Geomatics and Information Science of Wuhan University (12): (. AdaBoost [J] (12): [14]GongJianyaYao HuangShenXin.Object-Orien- [15]LiJianning.MultivariateAnalysisandItsApplica- tionin HigherEducationResearch[M].Hefei:An- huiuniversitypress2009(. [M]. : 2009) ( 1256 )

6 OntheStoreStrategyofSmalSpatio-TemporalData FilesinCloudEnvironment XIONG Lian 1 XU Zhengquan 1 WANG Tao 1 GU Xin 1 1 StateKeyLaboratoryofInformationEngineeringinSurveyingMappingandRemote SensingWuhanUniversityWuhan430079China Firstauthor:XIONGLianPhDcandidatespecializesin Webdatamininganddistributedstorage. xionglian302@163.com Foundationsupport:TheNational973Program ofchinano.2011cb302306;thenationalnaturalsciencefoundationofchinano 檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪檪 ( 1240 ) InteligentMonitoringforDenseCrowdUsingaFisheyeCamera HU Xuemin 1 ZHENG Hong 2 GUO Lin 1 1 SchoolofPhysicsandElectronicsHubeiUniversityWuhan430062China 2 SchoolofElectronicInformationWuhanUniversityWuhan430072China Abstract:Internetapplicationscreatemassivesmalspatio-temporaldatafilesincloudenvironments. Thereforeamethodaimingtoraisetheprocessingeficiencyofsmalfilesin HDFS;adatascheme combininguseraccessanddatafeaturesisproposed.thisschemeregardsauseraccessstreamasfile requestsequenceandconstructsacharacteristicsequencebyspatio-temporalatributeextraction.a characteristictemplateofdiferentuseraccesspaternsisformedwhenanalyzingthecharacteristicse- quencebytemplatematching.thenmerger-relatedfilesareanalyzed.experimentalresultsshowthat ourschemeimprovesthestorageeficiencyforsmalfilesandalsodecreasesnetworkapplicationre- sponsetimes. Keywords:cloudstorage;smalfile;spatio-temporaldata;featureextraction;templatematching Abstract:Crowdmonitoringindensecrowdscenehasbecomeanimportantanddificulttopicinthe fieldofautomaticsurveilancesystem.a densecrowd monitoringapproach based on perspective weightmodelforfisheyeimagesisproposedinthispaper.experimentalresultsshowthatthepro- posedapproachisefectiveandfeasiblefordensecrowdmonitoringinindoorscenes. Keywords:densecrowd;perspectiveweightmodel;densityestimation;crowdcounting Firstauthor:HU XueminPhDspecializesinimageprocessingmachinevisioninteligentcomputingandinteligentsystem. huxuemin2003@163.com Foundationsupport:ResearchFundofHubeiUniversityNo ;ScienceandTechnologyResearchProgramfortheYoungTal- entsofhubeieducationcommiteeofchinano.q

NP [11] 7) 4)~ 6) ( ) : Kirk- patrick [26] 1 Fig.1 TextureMergingSketch Map ) 2) 3) 1 4) 5)

NP [11] 7) 4)~ 6) ( ) : Kirk- patrick [26] 1 Fig.1 TextureMergingSketch Map ) 2) 3) 1 4) 5) 40 3 2015 3 GeomaticsandInformationScienceofWuhanUniversity Vol.40No.3 March2015 DOI:10.13203/j.whugis20140231 :1671-8860(2015)03-0347-06 1 1 1 1 430079 : : :P208TP751 :A ISOMAP [1] [2] [25] [3] [4-10]