Hand Segmentation from Depth Image using Anthropometric Approach in Natural Interface Development

Abstract

Abstract—Hand gestures are often used as natural interface between human and robot. To acquire hand gesture from a captured image, hand segmentation procedure is performed. In this manuscript, a method for hand image segmentation from depth image is proposed. This method uses image thresholding technique to obtain human image part from a depth image. The threshold level is obtained by analyzing human posture dimension (anthropometry). By finding the centroid of the human image, left and right regions of human body can be separated. Assuming that each region has a hand image and the hand is positioned in front of its body, both hand images can be located. Hand segmentation is started by using an anthropometric data of hand pose. The data is used to compute the color values that represent the depth of hand in each image region. Thus, the acquired values are used as threshold for each image region. The thresholding operation resulted in completely segmented hand images. This proposed method has low computation time and works well when the basic assumptions are fulfilled. Index Terms—anthropometric, depth image, gesture recognition, hand segmentation, natural interface, threshold. 1