Articulated hand tracking from single depth images using Gaussian Swarm Filtering

Abstract

Articulated hand tracking from video sequences is a challenging task which is often addressed in a particle filter framework. As it is difficult to perform dense sampling in a high-dimensional hand state space, the traditional particle filter can't track articulated hand motion well. In this paper, we propose a new algorithm which combines an improved Gaussian particle swarm optimization (Gaussian PSO) with a particle filter and use the new algorithm, termed Gaussian Swarm Filtering, to track articulated hand motion from single depth images obtained by a Kinect sensor. The improved Gaussian PSO is employed to move the particles towards the promising areas in the state space based on the newest observation. By using the depth information as the only input, our method is immune to background and illumination changes. An implementation of the proposed method is developed with OpenSceneGraph (OSG). Experiments based on synthetic data and real image sequences are both performed for evaluation. The results show that the proposed method is accurate and robust for articulated hand motion tracking.