Human action recognition using tensor principal component analysis
Human action can be naturally represented as multidimensional arrays known as tensors. In this paper, a simple and efficient algorithm based on tensor subspace learning is proposed for human action recognition. An action is represented as a 3th-order tensor first, then tensor principal component analysis is used to reduce dimensionality and extract the most useful features for action recognition. So the spatial and temporal correlations of the action are preserved. After then, a nearest neighbor classifier based on tensor distance is used to recognize action, in other words, measuring the similarity between actions using tensor distance in tensor subspace. The proposed method is assessed by using a public video database, namely Weizmann human action data sets. Experimental results reveal that the proposed method performs very well on that data sets, and robustness test has been carried out to testify the effectiveness.
Author's Name: Sun, M., Liu, X., Wang, S., Liu, Y., Zhou, C.
Volume: Volume 8
Issues: Issue 24
Keywords: Human action recognition, Tensor principal component, Tensor subspace learning