3D Body-Part Tracking of Two Persons Using a Hierarchical Body Model

Abstract

This paper presents a framework for hierarchical 3D articulated human body-part tracking of two persons. We introduce a Hierarchical Annealing Particle Filter (H-APF) algorithm which uses a Hierarchical Human Body Model (HHBM) in order to perform an accurate body pose estimation of several people. The human poses are presented in a high-dimensional space. The proposed method applies a nonlinear dimensionality reduction of this high-dimensional pose space to the low-dimensional latent spaces combined with the dynamic motion model and the hierarchical body model. The improved annealing approach is used for the propagation between different body models and sequential frames. The tracking algorithm generates trajectories in the latent spaces which provide low-dimensional representations of body poses observed during the motion. The approach was illustrated on HumanEvaI and HumanEvaII datasets as well as on several other datasets with different motions and proved to be effective and robust. A comparison with other methods and the error evaluation are provided.