Human Action Recognition in Videos Using an Effective Combined Method

Abstract

The local and global features have attractive characteristics and they are complementary each to other therefore, utilizing these features as a combination in the human action recognition system leads to increase the recognition accuracy. In this paper an effective human action recognition method has been proposed based on the combined histograms of local and global features. The local features are provided using radon transform descriptor, while, the global features are provided using contourlet transform descriptor. The cosine distance measure has been used to compute the distance between the test and training frames of the video sequences. The extensive experiments carried out on two standard video databases indicated that the proposed method achieved high recognition rates. Moreover, the results of the comparison carried out between the proposed method and some other human action recognition methods refer that the proposed method outperforms these methods.Keywords: Human action recognition, local and global features, radon transform, contourlet transform, cosine distance measure.