Human pose estimation using a mixture of Gaussians based image modeling

In this paper, we propose an approach toward body parts representation, localization, and human pose estimation from an image. In the image, the human body parts and a background are represented by a mixture of Gaussians, and the body parts configuration is modeled by a Bayesian network. In this model, state nodes represent pose parameters of an each body part, and arcs represent spatial constraints. The Gaussian mixture distribution is used to model the prior distribution for the body parts and the background as a parametric model. We estimate the human pose through an optimization of the pose parameters using likelihood objective functions. The performance of the proposed approach is illustrated on various single images, and improves the human pose estimation quality.