An Gaussian-Mixture Hidden Markov Models for Action Recognition Based on Key Frame

Abstract

When using Gaussian-Mixture Hidden Markov Models (GMM-HMM) for action recognition, the accuracy of recognition is greatly improved. However, the number of Gaussian Mixed Models (GMM) and Hidden Markov Models (HMM) classifications needs to be defined. In this paper, we propose a key frame-based GMM-HMM motion recognition method. Specifically, we use the minimum reconstruction error method to determine the number of key frames (KFN). Then, we set the number of GMM and HMM classifications to be KFN. In the end, we use experiments with three different dataset to test our method.