A CRF Based Model for Learning High Level Behaviors of the Elders in Household Environment

Abstract

A CRF model is proposed to extract the high level behaviors based on the position information of the elders in household environment. Firstly, the slide time window is used to segment the time-order position data which is then transformed into the position-based information including position region No., speed and direction. Then, we take the information as the observation data to construct the CRF model, learn the model parameters, and achieve to extract the high level behavior state sequences which can be applied to detect the abnormality of the elders. The experiment results show that the proposed method can identify the behavior states with a good rate of identification accuracy.