Vision Based Real-time Gesture Tracking And Recognition And Its Application In Human-Computer Interaction Techniques

With the development of human-computer interaction techniques, variety of new interaction techniques are springing up.Vision-based gesture recognition technique become one of key techniques of human-computer interaction because of its some notable advantages, e.g. more natural in human daily life. Vision-based gesture recognition is an active research topic in the fields of image processing, pattern recognition, computer vision and others. Vision-based gesture recognition has wide applied value. For example,(1)the control of AI in VR, (2)robot control, (3)multi-pattern interface in VR system.At present,the research about gesture recognition is mostly static based. However,the research of real-time and dynamic gesture tracking and recognition is lacked.This thesis innovatively researched vision based hand tracking and gesture recognition in complex background. With the steady perfection of the experiment, this technique will promote the application of gesture recognition and make human-computer interaction more natural and convenient.Firstly, a new method based on skin abstraction combining motion information is proposed by this thesis in the stage of gesture division. The YCbCr color space and Gaussion Function based skin color modeling are used to abstract the skin color.Then by using the image subtraction operation the skin-color like background and face area can be eliminated.The algorithm can make the gesture division in complex background accurate.Then in the part of gesture tracking,this thesis applied the CAMSHIFT algorithm combining Kalman filter algorithm to find the research window and the gesture postion. CAMSHIFT algorithm is applied to estimates the center and size of tracking window.Then the Kalman filteris used to predict the hand region centroid.This algorithm improved the tracking efficiency and ensured the real-time performance.Finally in the step of gesture recognition,the method of model matching using Hausdorff distance was used to recognize the two defined gesture. To validate the method above,a vision-based gesture interaction systerm was designed which can implement the cursor moving and click.Through testing the system has high tracking efficiency and can recognize the two defined gesture effectively.