Human pose estimation in visual computing
The present invention discloses a method of estimating human pose comprising: modeling a human body as a tree structure; optimizing said tree structure through importance proposal probabilities and part priorities; performing foreground detection to create image region observation; and performing image segmentation to provide image edge observations.
1. A method of estimating human pose comprising:
modeling a human body as a tree structure;
optimizing said tree structure through importance proposal probabilities and part priorities;
performing foreground detection to create image region observation;
performing image segmentation to provide image edge observations;
changing and propagating said part priorities, part dynamic probabilities, and state dynamic probabilities; and
running local optimization under data-driven Markov chain Monte Carlo (DDMCMC) framework.
2. The method of claim 1 wherein said tree structure comprises 3 levels.
3. The method of claim 2 further comprising:
sampling torso first to initialize body states after modeling said human body.
4. The method of claim 3 further comprising:
sampling other body parts to find local extremum after sampling torso first.
5. The method of claim 1 further comprising:
choosing dynamic of jump to represent large changes of part states.
6. The method of claim 5 further comprising:
choosing dynamic of diffusion to represent small changes of part states.
7. The method of claim 6 further comprising:
determining whether to accept new states or not by Metropolis Hasting approach after running said dynamic.