Efficient posture estimation using the particle filter

We address the following problem: estimate the posture vector sequence from a monocular image feature vector sequence. Its solution can be used to realize visual interfaces between humans and intelligent machines. We take the state space approach, that is, the posture estimation is formulated as the state estimation. We propose a cyclic motion model (CMM) whose state variable is the phase of a motion, and estimate the postures using a particle filter. In addition, we extend CMM so that the posture estimation is possible from image sequences with camera angle change.