Supervised Neighborhood Topology Learning for Human Action Recognition

Supervised manifold learning has been successfully applied to human action recognition. With the class label information, the recognition performance can be improved. However, the learned manifold may not be able to well preserve the local structure which reflects temporal information of an action. To overcome this limitation, this paper proposes a new supervised manifold learning algorithm namely supervised neighborhood topology learning (SNTL) for human action recognition. SNTL is based on the framework of locality preserving projection (LPP). Different from LPP, SNTL constructs the adjacency graph with a topology defined in a supervised manner, which not only separates data points from different actions but also preserves the local structure of data points from the same action. With the advantage of locality preserving property in the framework of LPP, SNTL provides good discriminant ability and preserves temporal information of each action contained in local structure. Weizmann human action database is used for evaluation. Experimental results show that the method achieves 95.56% recognition accuracy.