| In the filed of computer vision, view-invariant action recongnition has become a hot issue. This direction has been profound significance in Artificial intelligence, Human computer interaction, security monitoring, and video editing. In this paper, the main research is about the building of the descriptor of the same action with multi-view points. We analysis the method of the description about the same action with multi-views in a video, and achieve the goal of view-invariant action recongnition.In this paper, we explore the invariance property of temporal order of action instances during action execution and utilize it for devising a new view-invariant action recognition approach. We utilize spatiotemporal features, feature fusion and temporal order consistency, in order to ensure temporal order during matching. We begin with extracting spatial-temporal interesting points (STIPs) form video sequences and then apply feature fusion to encapsulate within class similarity for the same view-points. We construct a feature fusion table to facilitate feature matching across different views for each action class. Based on global temporal order constraint and number of matching features, the action matching score is then calculated. Finally, the action label of the class with the maximum value of the matching score is assigned to the query action. The action semantic of the label representing is considered to be the meaning of recognized action.At last, this paper test the algorithm with lots of experiments. We perforate a large of number of experiments to evaluate the algorithm with multiple view inria xmas motion acquisition (IXMAS) action datasets. The results of experiments show this algorithm has better performance in view-invariant action recognition. That is to say that the algorithm on temporal order invariance for view-invariant action recognition has an effective and ability to recognize the same actions with multi-view points in a vide, which exhibits superior robustness. |
PDF Full Text Request