Side view pose estimation of human from images using prior knowledge

A new framework is proposed to estimate the side view pose of human from still images. The framework applies pose prior knowledge into the pose estimation process of pictorial structure and get a good performance. This paper has three main research contributions: 1.Proposing a new silhouette extraction method based on color clustering analysis and foreground highlighting; 2.Proposing a hierarchical pose classification method to solve the pose-duality problem and a learning method for pose prior knowledge from training set; 3. Proposing the pose prior knowledge restricted image parsing process to make up the image information loss due to the inter-occluded between body parts and improve the estimated performance. Experiment results on estimating the human walking pose prove that the framework can evidently reduce the interference from background clutter and is able to estimate the reliable side view pose of human from still images.