Human Pose Estimation Based on Evidence Supporting and Sub-graph Pruning

The proof of human parts has an imperative effect on pose evaluation, and can be effortlessly confused with difficult background due to indefinite part detector. This paper circumvents this predicament by performing a proof supporting approach, where each part also receives confidence from its neighborhood which uses the outline information between connect parts and mitigates the risk of being blindly pruned. Firstly, we present a novel part learning technique to deal with the shape deformation and image misalignment. Our technique can identify human parts candidates competently and effectively based on the numerous instance learning frameworks. Next, we recommend a proof supporting technique which utilizes the mutual context information of connected parts to boost the confidence of human parts. By using this approach, it prevents the human parts with feeble proof to be pruned blindly. Finally, we propose a sub-graph pruning technique, which process all the parts hierarchically in order to decrease the computational intricacy, to minimize the state space of human parts. Experimental results on three public datasets indicate that our proposed technique can deal with the articulated human pose estimation competently and get the superior results.