A hybrid approach to 3D arm motion tracking

Abstract

This paper presents a hybrid approach to 3D arm motion tracking for tele-rehabilitation applications. A particle filter (PF) algorithm is adopted in the proposed system to fuse data from inertial and visual sensors in a probabilistic manner. Multi-modal distributions of system states are propagated based on a 'factor sampling' technique. To avoid the problem of particle degeneracy in conventional PF algorithms, two strategies are adopted in our system, namely state-space pruning and an arm physical geometry constraint. Experimental results show that the proposed PF framework outperforms other fusion methods and tracking results are accurate in comparison to the ground truth provided by a commercial mark-based motion tracking system.