This strong interest is driven by a wide spectrum of promising app lications in many areas such as virtual reality, smart surveillance, percep tual user interface, content-based image storage and retrieval, athletic performance analysis, etc. Bymaking use of hidden Markov models (HMMs) , the recognition of human motion image is studied. This includes selecting new method of extracting eigenvector and designing algorithms of HMM training and recognizing.Experimental results show that thismethod is practicable for the human activities analysis.