RGB-D camera-based hand shape recognition for human-robot interaction

Abstract

Hand is the most popularly used tool for human-robot interaction. Therefore, this paper proposes a Kinect-based hand shape recognition method for human-robot interaction. Kinect can capture color and depth images simultaneously and its SDK provides functions to track the human skeleton. Therefore, the proposed method can detect hands robustly by using the skeleton and depth information. In results, it can recognize various hand shapes based on contour analysis with a high recognition rate (95% on average) and works in real-time (over 30 frames/sec).