3D hand skeleton model estimation from a depth image

Chin-Yun Fan, Meng-Hsuan Lin, Te-Feng Su, Shang-Hong Lai, Chih-Hsiang Yu

Abstract: In this paper, we present an algorithm for estimating 3D hand skeleton model from a single depth image based on the Active Shape Model framework. We first collect a large amount of training depth images, representing all articulated hand shape variations, and a set of hand joint points are labeled on these depth images. To accommodate the wide variations of hand articulations, we represent the hand skeleton model with multiple PCA models that are learned from the training data. In the search stage, we iteratively compute the translation and rotation from the hand depth information and fit the 3D hand skeleton model with the multiple PCA models. In addition, we modify the model fitting procedure to handle the partial occlusion problem when only some fingers are visible. In our experiments, we demonstrate the proposed algorithm on our hand depth image datasets to show the effectiveness and robustness of the proposed algorithm.