简介:ThetraditionalorientedFASTandrotatedBRIEF(ORB)algorithmhasproblemsofinstabilityandrepetitionofkeypointsanditdoesnotpossessscaleinvariance.Inordertodealwiththesedrawbacks,amodifiedORB(MORB)algorithmisproposed.Inordertoimprovetheprecisionofmatchingandtracking,thispaperputsforwardanMOKalgorithmthatfusesMORBandKanade-Lucas-Tomasi(KLT).ByusingKalman,theobject’sstateinthenextframeispredictedinordertoreducethesizeofsearchwindowandimprovethereal-timeperformanceofobjecttracking.TheexperimentalresultsshowthattheMOKalgorithmcanaccuratelytrackobjectswithdeformationorwithbackgroundclutters,exhibitinghigherrobustnessandaccuracyondiversedatasets.Also,theMOKalgorithmhasagoodreal-timeperformancewiththeaverageframeratereaching90.8fps.