3D Articulated Human Body Tracking by Particle Filter

For sophisticated background, a human body tracking algorithm using particle filter based on 3D articulated body model is introduced. First, a high-fidelity biomechanical upper body model was built, which is accurate for representing varies complicated human poses and is simple to be developed. Then a sequence of images was obtained by using a stereo camera. After calibration, verification and background subtraction, the information resources of calculated depth map, foreground silhouette and arms skeleton were chosen to construct the likelihood function. The state vectors describing the human pose were computed by fitting the articulated body model to observation human using particle filter. The torso and arms were tracked separately so that the computational complexity and the number of particles could be reduced. Experimental results show that the proposed algorithm can rapidly and accurately track human body with different poses, different person under different illumination conditions.