Articulated-model based upper-limb pose estimation

For human motion analysis and recognition, the current trend has extended from the tracking and recognition of hand gesture to pose estimation, body motion analysis and body language recognition and from the appearance based to 3D human model based. Since the major part of the human body language is performed by the upper limbs, and also for the sake of reducing the complexity, this paper only models the upper limbs on human and uses articulated model to reconstruct the 3D pose. It has been proved that it is feasible and reliable to detect the hands and eyes by means of stereo vision. Further, their 3D coordinates can be estimated and serve as approximate constraints. Based on camera perspective model in the font view image, integrating the approximate constrains and the skeleton length scales, this paper shows the precise 3D coordinates of each joint can be obtained iteratively from coarse to fine by hierarchical quasi-Voronoi.