Layered hidden Markov models for real-time daily activity monitoring using body sensor networks

This paper presents an inferring and training architecture for the long-term and continuously monitoring daily activities using a wearable body sensor network. Energy efficiency and system adaptation to subjects are two of the most important requirements of a body sensor network. This paper proposes a two-layered hidden Markov model (HMM) architecture for in-network data processing to achieve energy efficiency and model individualization. The bottom-layer HMM is used to preprocess the sensor readings locally at each wireless sensor node to significantly reduce the amount of data to be transmitted. The top-layer HMM is utilized to find the activity sequence from the result of this local preprocessing. This approach is energy efficient in that only the results of the decoding procedure in each node need to be transmitted rather than the raw sensor readings; therefore, the volume of transmitting data is significantly reduced.