Study of Different Image Features Used for Recognizing Unknown Person's Activity Based on Tensor Decomposition Approach

摘要

This paper proposes a method for recognizing an unknown person' action from a video sequence, where the unknown person is defined as a person who is not included in the database (tensor) used for the recognition process. We have developed a tensor decomposition based method, but that method used only motion capture data. The main motivation of this paper is to replace the motion capture data by image feature data and clarify which image features are useful. This paper tests three image features as well as their combinations. Experimental results using synthesized image sequences present the effectiveness of the method and clarify the useful image features.