Initial pattern library algorithm for human action recognition

Abstract

Human action recognition is currently one of the most active research topics in society management, including human moving detection, human moving classification, human moving tracking, and activity recognition and description. In this paper, we have proposed a new classifying and sorting initial pattern library algorithm for human action recognition. First, we classify the training vector set to two subsets by vector variance. Secondly, sort the subsets to put the similar pattern vectors together. Last, select some number of pattern vectors from the sorted subsets to form the initial pattern library. This new initial pattern library is tested by self-organizing maps (SOM) algorithm. Experimental results in image recognition show that this new initial pattern library algorithm is better than the common random sampling initial pattern library.