Human Action Classification Research With Video Data

In recent years, human behavior analysis is one of the most concerned topics in the field of computer vision. Human action in video sequences can be seen as silhouettes of a moving torso and protruding limbs undergoing an articulated motion. In accordance with the order of behavior analysis of human action, we analyzed and studied the human action detection, feature extraction and classification of human actions.In this paper, we use the method based on the 3D Poisson equation to extract the features of human actions. The local space-time saliency, action dynamics, shape structure, and orientation can be extracted by utilizing properties of the solution to the Poisson equation. In this paper, we use the space-time saliency and the space-time orientation to represent the local features of the silhouette sequence. We get the global features by calculating the 3D Zernike moment features of the overall silhouette sequence.This paper integrates Bayesian classifier and Adaboost classifier. We use the Bayesian-based Adaboost classifier to classify the human action. Experiments were carried out with the Weizmann video database. We separated the database into several parts. And we carried out several experiments with some combinations of the parts. Finally, we analyzed the results and compared them with other methods.Our experimental results show that this method can effectively extract the features of human action. And we can get better classification results on the bending, hopping, vertical jumping, side walking, walking, waving and some other actions.