Hand gesture recognition based on skeleton of point clouds

In this paper, we present a method of recognizing hand gestures in the form of point clouds recorded by Kinect sensor. Firstly, through Laplacian-based contraction and further processing, we extract skeleton points from point clouds of hands. Then, we apply a novel partition-based descriptor and corresponded algorithm to classify these skeletons and, taking one step further, to recognize gestures. In the process of recognition, the issue of scale variant and rotation variant are solved. Finally, to test and verify performance of our method, we design a series of experiments. Experimental results proved both its accuracy and robustness. Besides, we believe the skeleton-based way of recognition owns potential for further exploration.