Research On Action Recognition Based On The Space-Time Interest Points

In recent years, action recognition is the popular research field in the computer vision, and it is very challenging problem. The main purpose of action recognition identify the action type automaticly from the video. It is the high-level semantics for computer, and for the development of computer intelligence has an important significance. The action recognition technology have many applications in many computer field ,including the interface between the computer and human, video security surveillance , actomotive video mark system , the multimedia search and so on.Action recognition based on the space time interest points is the main research way in the recent of action recognition which is using some classic image processing techniques extended to the three-dimensional motion video processing. This method includes detecting the space time interest points ,getting the local features, making the bag of words model and training and predicting by the support vector machine. In this paper, the method combinates the some recent local decription such as the Histogram of Oriented Gradients, Histogram of Optical Flows with the Shape of Gaussians to the new description. And in this theory, the action recognition have the state-of-the-art performance.In the last of paper, action recognition based the space time points and Shape of Gaussians with other decription has the state-of-the-art performance, and this ideas inspire the action recognition approach by a combination with varity of decriptions.