Robust hand posture recognition based on RGBD images

Abstract

This paper proposes a robust hand posture recognition system based on RGBD images. While much research has focused on human body posture recognition, this work investigates skeleton-free hand detection, tracking and posture recognition. This work consists of two different parts. In the first part, we utilize random forest to get pixel detection of hand and mean-shift to track hand based on RGBD images. In the second part, we implemented extraction of different features and RBF support vector machine to recognize multiple hand postures. This system has two advantages: it is skeleton-free and works in wider area; it is more robust by combining depth and color features. At last, we use the posture recognition system to control robots in virtual reality platform.