Real-time fingertip tracking and detection using Kinect depth sensor for a new writing-in-the air system

Abstract

We propose a real-time finger writing character recognition system using depth information. This system allows user to input characters by writing freely in the air with the Kinect. During the writing process, it is reasonable to assume that the finger and hand are always holding in front of torso. Firstly, we compute the depth histogram of human body and use a switch mixture Gaussian model to characterize it. Since the hand is closer to camera, a model-based threshold can segment the hand-related region out. Then, we employ an unsupervised clustering algorithm, K-means, to classify the segmented region into two parts, the finger-hand part and hand-arm part. By identifying the arm direction, we can determine the finger-hand cluster and locate the fingertip as the farthest point from the other cluster. We collected over 8000 frames writing-in-the-air sequences including two different subjects writing numbers, strokes, pattern, English and Chinese characters from two different distances. From our experiments, the proposed algorithm can provide robust and accurate fingertip detection, and achieve encouraging character recognition result.