An approach based on wavelet analysis and hidden markov models for behavior understanding

This paper proposes a novel parallel structure and an individual multibehavior module which is based on hidden Markov models (HMM) to extract behavior features. In the parallel structure, the wavelet de-noising method is employed to preprocess data and provide the robust training data. Then, the individual multi-behavior module is built to analyze multi-sensor signals to obtain the behavior features. The experimental results show that this method is useful for behavior understanding such as sleeping, for which the matched probability is up to 90%.