Effective search space reduction for human pose estimation with Viterbi recurrence algorithm

In this paper, an efficient algorithm for estimating human pose in static images is presented, which is based on pictorial structure model and Viterbi recurrence algorithm. Our algorithm mainly solves three problems in the process of estimating human pose: 1) for overcoming the influence of illumination change and local deformation, a new part appearance model based on HOG feature and SVM is presented; 2) for reducing search space, a new approach using location prior and matching threshold is presented, which can also achieve increasing the rate of convergence and improving the accuracy of human pose estimation; 3) an inference algorithm using Viterbi recurrence algorithm is designed. Experiments results show this new algorithm is more efficient.