Posture estimation by Bayesian Network with Belief Propagation

A Bayesian Network was proposed to estimate human body posture in three dimensional using a probabilistic inference way. In this study, to represent and reconstruct the motion of human body, a three dimensional rigid links model which consists of bones and joints was built. Belief Propagation algorithm was employed to implement Bayesian probabilistic inference for the estimation task. Based on the human body model and Bayesian network, a simulation of walk and falling down was conducted to validate the proposed method. The orientation and position of human body were estimated by the graphic network. The simulation results showed that the proposed method was accurate for human body posture estimation. Further, the proposed method may have potential for human motion prediction.