View-invariant human action recognition using exemplar-based Hidden Markov Models

Abstract

An exemplar-based Hidden Markov Model is proposed for human action recognition from any arbitrary viewpoint image sequence. In this framework, human action is modelled as a sequence of body poses (i.e., exemplars) which are represented by a collection of silhouette images. The human actions are recognized by matching the observation image sequence to predefined exemplars, in which the temporal constraints were imposed in the exemplar-based Hidden Markov Model. The proposed method is evaluated in a public dataset and the result shows that it not only reduces computational complexity, but it also is able to accurately recognize human actions using single cameras. © 2009 Springer-Verlag Berlin Heidelberg.