Robust 3D arm tracking from monocular videos

Abstract

In this paper, we present a robust method to tackle the ambiguities in 3D arm tracking, especially those introduced by depth change (distance of the arm from the camera), and arm rotation about humerus (upper arm bone). In a particle filter framework, the arm joint angle configurations are monitored and the occurrences of the ambiguous arm movements are detected. Inverse kinematics is applied to transfer invalid joint angle configurations from unconstrained movement space into constrained space. Experimental results have demonstrated the efficacy of the proposed approach. © Springer-Verlag Berlin Heidelberg 2005.