A New Method of Feature Description for Human Action Recognition

Human action recognition is an important task for intelligent video surveillance. The first step of the algorithm for the task are feature extraction and description. The result of action recognition is strongly depend on the accuracy of this step. Many method such as SIFT, HOG, Optical flow are proposed for a robust feature detection. However, it is still difficult to realize robust detection for many scenes by these method. In this paper, we adopted a method which is firstly used in action recognition. We also proposed a new method to describe feature. The results show that our method for feature extraction and description could improve the accuracy of action recognition.