Abstract: Human body is decomposed into several local component templates connected with tree structure by articulated pose estimation with mixture-of-parts, and the corresponding full score is used to estimate body pose. Such a method fails in precisely estimating human pose due to the defects of tree models when human has the complicated pose of body. To address this problem, pose estimation using limbs adjustment with mixtures-of-parts models was proposed. And pixel discrete direction, local characteristic functions blending full score and pixel discrete direction and other attributes, and pose estimation using limbs adjustment based on local characteristic functions of articulated position were proposed, which is better than the method based on articulated pose estimation with mixtures-of-parts.

Key words: mixture-of-parts, pose estimation, limb adjustment, pixel discrete direction, full score