Human Action Recognition: A Novel Application of the Stochastic Detector Enriched by Body-parts Information

摘要

In this paper, we investigate the problem of human action recognition from the video. And a pose-to-action hierarchical recognition step is adopted. Different from previous works which define some particular distances in the Euclidean space to measure the difference between any two human poses, we propose a new Probabilistic Pattern Distance (PPD). This measurement fully considers the ambiguity and the rich information contained in a human body. Then, Hidden Markov Model (HMM) will be used to recognize the human action. Final experimental results demonstrate that our model can effectively improve the action recognition performance, especially when a stable human silhouette image can not be obtained.