摘要
ThetraditionalorientedFASTandrotatedBRIEF(ORB)algorithmhasproblemsofinstabilityandrepetitionofkeypointsanditdoesnotpossessscaleinvariance.Inordertodealwiththesedrawbacks,amodifiedORB(MORB)algorithmisproposed.Inordertoimprovetheprecisionofmatchingandtracking,thispaperputsforwardanMOKalgorithmthatfusesMORBandKanade-Lucas-Tomasi(KLT).ByusingKalman,theobject’sstateinthenextframeispredictedinordertoreducethesizeofsearchwindowandimprovethereal-timeperformanceofobjecttracking.TheexperimentalresultsshowthattheMOKalgorithmcanaccuratelytrackobjectswithdeformationorwithbackgroundclutters,exhibitinghigherrobustnessandaccuracyondiversedatasets.Also,theMOKalgorithmhasagoodreal-timeperformancewiththeaverageframeratereaching90.8fps.
出版日期
2016年02月12日(中国期刊网平台首次上网日期,不代表论文的发表时间)