Estimating 3d human pose from single images using iterative refinement of the prior

This paper proposes a generative method to ex-tract 3D human pose using just a single image. Un-like many existing approaches we assume that accu-rate foreground background segmentation is not pos-sible and do not use binary silhouettes. A stochastic method is used to search the pose space and the poste-rior distribution is maximized using Expectation Max-imization (EM). It is assumed that some knowledge is known a priori about the position, scale and orienta-tion of the person present and we specifically develop an approach to exploit this. The result is that we can learn a more constrained prior without having to sacri-fice its generality to a specific action type. A single prior is learnt using all actions in the HumanEva dataset [9] and we provide quantitative results for images selected across all action categories and subjects, captured from differing viewpoints. 1