Discovering Commonness and Specificness for Human Action Recognition

Human action recognition remains a challenging problem though having been intensively researched for decades. Recently, many sparse coding based approaches have been proposed to advance the progress in this research field. However, most of these approaches aim to learn a more discriminative dictionary by incorporating various regularization terms so that sparse codes are more representative for better recognition performance. Instead, in this paper, we propose a novel discriminative dictionary learning method which recognizes the commonness and specificness among different action classes. That is, we aim to obtain a universal dictionary which consists of two parts, a shared dictionary for all action classes and a set of class-specific dictionaries. As a result, inter-class differences can be better characterized with sparse codes obtained from the class-specific dictionaries. In addition, group sparsity constraint is utilized to ensure that similar descriptors of the same action class have similar sparse codes and locality constraint is utilized to ensure data locality. The experimental results on the popular UCF sports dataset demonstrate that our proposed approach outperforms the state-of-the-art of related methods.