This paper presents a framework for 3D articulated human body tracking and action classification. The method is based on nonlinear dimensionality reduction of high dimensional data space to low dimensional latent spaces. Human body motion is described by a hierarchy of low dimensional latent spaces which characterize different groups of body parts. We introduce a body pose tracker thats uses the learned mapping function from latent spaces to body pose space. The algorithm initially makes a rough estimation of body pose and then improves it using the Hierarchical Human Body Model. The trajectories in the latent spaces provide low dimensional representations of body pose sequences representing a specific action type. These trajectories are used to classify human actions. The approach is illustrated on the HumanEval and HumanEvall datasets, as well as on other datasets. A comparison to other methods is presented.
展开▼
