A Human Action Recognition Algorithm Based on Semi-supervised Kmeans Clustering

Abstract

This paper proposes a new method of semi-supervised human action recognition. In our approach, the motion energy image(MEI) and motion history image(MHI) are firstly used as the feature representation of the human action. Then, the constrained semi-supervised kmeans clustering algorithm is utilized to predict the class label of unlabeled training example. Meanwhile the average motion energy and history images are calculated as the recognition model for each category action. The category of the observed action is determined according to the correlation coefficients between its feature images and the pre-established average templates. The experiments on Weizmann dataset demonstrate that our method is effective and the average recognition accuracy can reach above 90% even when only using very small number of labeled action sequences.