Identification, Based On The Overall Characteristics Of Human Motion

At the current study,' the research on human body movement recognition has already made a great of progress. But the result is not so ideal when dealing with the data from Weizmann Database which includes multi-cycle action video. This paper provides a new method which is based on global features to solve this problem.First of all, moving targets are extracted using the method of background subtraction. Second, we generate two static models which are motion-energy-image (MEI) and motion-history-image (MHI), and make MEI and the single frame image as the origins of movement. Third, Zernike moments are used as global features. Fourth, a new approach based on Bag of words is presented for characteristics clustering. The clustering algorithm is used to generate the key point, which is the standard of classification and after that we will get the histogram of feature classification. Fifth, the fusion of multiple features is used to improve the correct rate of recognition. Finally, Support Vector Machine (SVM) is utilized as the classifier to recognize the movements.Six sets of experiments were designed. The experimental results on six sets of experiments demonstrate that the approach in this paper can perform very well.