简介：Thispaperpresentsahumanactionrecognitionmethod.Itanalyzesthespatio-temporalgridsalongthedensetrajectoriesandgeneratesthehistogramoforientedgradients(HOG)andhistogramofopticalflow(HOF)todescribetheappearanceandmotionofthehumanobject.Then,HOGcombinedwithHOFisconvertedtobag-of-words(BoWs)bythevocabularytree.Finally,itappliesrandomforesttorecognizethetypeofhumanaction.Intheexperiments,KTHdatabaseandURADLdatabasearetestedfortheperformanceevaluation.Comparingwiththeotherapproaches,weshowthatourapproachhasabetterperformancefortheactionvideoswithhighinter-classandlowinter-classvariabilities.IndexTermsBag-of-words(BoWs),densetrajectories,histogramofopticalflow(HOF),histogramoforientedgradient(HOG),randomforest,vocabularytree.