VIDEOEVENT DETECTION USINGICAMIXTURE HIDDEN MARKOV MODELS JianZhou
Toronto Office, Xiao-Ping Steven Zhang · 2006
Inthis paper, aframework that combines feature extraction, modellearning, andlikelihood computation, ispresented for video event detection. First, theindependent component analysis(ICA)isapplied totherawfeature space toextract the spatial features. Then, aframework based onICAmixture hidden Markovmodels (ICAMHMM)isusedtoexploit the spatial andtemporal characteristics ofthetraining data. After themodelislearnt, thelikelihood foragiven video sequence iscomputed andthenusedtoclassify thevideo into asemantic event. Golfvideo sequences areusedforsimulations. The results showthat theproposed method caneffectively detect semantic video events. IndexTerms-Videoevent detection, independent component analysis (ICA), hidden Markovmodel, ICAmixture, feature extraction, semantic analysis