Research On Vision-Based Hand Gesture Recognition

Gesture recognition, as one of the most natural way of Human Computer Interaction, is one of hottest research field in Computer Vision. Lately a lot of studies focus on this domain trying to figure out an efficient way to solve the problem that tracking hand and recognizing gesture in real-time. This paper also focuses on this field, trying to using WebCam and Microsoft Kinect as input devices to recognize gestures. There are three parts in this paper:(1)Robust and real-time hand tracking using WebCam; (2)Posture recognition using Kinect; (3)Gesture recognition using Kinect.This paper proposed a robust method to tracking hand in complex background. Frist we use certain posture to launch up tracking procedure. Then generate probability graph using Bayes skin color model. After that we using probability graph to calculate local velocity and global velocity and update tracking window using these information. Then we use CAMSHIFT algorithm to refine the tracking region. Finally we update tracking information for the next tracking loop. Compare to previous works our method can get a better result in clutter scene in which exists big skin-like areas.This paper also proposed a method to recognition posture using Kinect. Firstly, we get hand region based on tracking result. Secondly, sampling feature from contour of hand region. Thirdly, we calculate Shape Context features for hand shape. Finally we determine the shape type using KNN algorithm. More than that, we also implemented an application which can recognize gesture which represents certain numbers.We also proposed a method to recognition gesture based on Kinect and HMM. First we discussed the choice of observation. Then we discussed how to detect start point and end point. Finally we introduced our application which can recognize characters written by bare hand in front of Kinect.