A Real-time Hand Postures Estimation Method

Abstract

Hand postures can be used to achieve friendly human-machine interaction (HMI) for the advantages of simple expressions, informative instructions and unconstrained operations. However, most previous postures estimation methods have problems such as poor real time performance, low estimation accuracy, being sensitive to the varying illumination and relying on marks on hands or special assistant devices. In this paper, we will introduce an efficient real-time method to estimate hand postures with an ordinary monocular camera. This method contains two stages, which are hand segmentation and postures recognition. Rather segment objects from the whole image, we locate the hand within a local area first and then adapt motion, color and contour informations to segment the hand region accurately. After that, template matching algorithm is used to recognize hand postures with fingertips and shape features. The experimental results demonstrated that the introduced method realizes high estimation accuracy and fast processing speed. Furthermore, the method is robust to varying illumination conditions.