With the successful introduction and popularization of Kinect, it has been widely applied in intelligent surveillance, human-machine interaction and human action recognition and so on. This paper presents a human action recognition based on multimodal information using the Kinect sensor. Firstly, the HOG feature based on RGB modal information, the space-time interest points feature based on depth modal information, and the human body joints relative position feature based on skeleton modal information are extracted respectively for expressing human action. Then, the three kinds of nearest neighbor classifiers with different distance measurement formulas are used to predict the class label for a test sample which is respectively expressed by three different modal features. The experimental results show that the proposed method is simple, fast and efficient compared with other action recognition algorithms on public datasets.
|