Dynamic Hand Gesture Recognition Using Depth Data and Kernel Density Estimation of Fingertip Angle Set

摘要

The lack of robustness and real-time ability under complex environment are the two common problems of dynamic hand gesture recognition. This paper presents a fast and stable algorithm based on Kinect depth data and kernel density estimation of fingertip angle set. In Kinect scene, 3D trajectory of dynamic hand gesture is extracted and recognized stably based on depth data and accurate hand gestures. By defining the fingertip angle set of hand shape, the kernel density estimation sequence of fingertip angle set (KDES-FAS) feature is extracted robustly. Finally, the dynamic hand gesture is recognized with the combination of 3D trajectory and hand shape. The extensive experiments demonstrate that the recognition method possesses better real-time performance and robustness for any dynamic hand gestures. The method proposed in this paper has both theory meaning and better value of application.