简介:Inrecentyears,accordingtotheneedofintelligentvideosurveillancesystemincreasingrapidlyinmetropolitancities,adesignbasedonS3C2440microprocessorandembeddedLinuxoperatingsystemisadoptedforreal-timevideotargettracking.However,itisverychallengingasembeddedsystemsusuallyaffordlimitedprocessingpowerandlimitedresources.Therefore,toaddressthisproblem,areal-timetrackingalgorithmusingmulti-featuresbasedoncompressivesensingisproposedandimplemented.Thealgorithmusesmultiplematrixastheprojectionmatrixofthecompressivesensingandthecompresseddateasthemultiplefeaturestoextractusefulinformationneededbytrackingprocess.FunctionsandlibrariesinOpenCVwhichweredevelopedbyIntelCorporationareutilizedforbuildingthetrackingalgorithms.Itistestedwithvariantvideosequencesandtheresultsshowthatthealgorithmachievesstabletrackingforthetargetmovedofthelightchanged.
简介:3Dobjectscanbestoredincomputerofdifferentdescribingways,suchaspointset,polyline,polygonalsurfaceandEuclideandistancemap.Momentinvariantsofdifferentordersmayhavethedifferentmagnitude.Amethodfornormalizingmomentsof3Dobjectsisproposed,whichcansetthevaluesofmomentsofdifferentordersroughlyinthesamerangeandbeappliedtodifferent3Ddataformatsuniversally.Thenaccuratecomputationofmomentsforseveralobjectsispresentedandexperimentsshowthatthiskindofnormalizationisveryusefulformomentinvariantsin3Dobjectsanalysisandrecognition.