Dynamic Hand Gesture Recognition For Home Service Robot

With the development of computer technology. The way that human interact with computer is developing towards humanity and simplicity. Natural user interface means users communicate with the computer in a natural way such as voice, touch, gesture and so on. Hand gesture is a natural and intuitive way of interpersonal communication.Gesture recognition based on vision is a key technology to realize the humanity of human-computer interaction. Kinect is a revolutionary product which provides a new way of human-computer interaction and expresses the idea of human-computer interaction more completely. It can overcome the influence of changing illumination. The Kinect could captures and tracks the body’s movements, gestures and sounds. In this thesis, we used the three-dimensional information provided by Kinect. We get the characteristic values by calculating the angle between the vectors formed by the skeletons. We proposed a KNN fast recognition based on the spherical region. The work of this paper mainly has two aspects.Firstly, we analysis the important joints of dynamic hand gesture recognition which determine the posture of dynamic hand gesture. After determining the key joints, we carry out feature extraction based on the three-dimensional information of the skeletons. We calculate the angle between Z axis and vectors which are composed of two important joints and the angle between X axis and the line which the two-joints-made vector projects in the plane of the XY axis. Then we considered the head joint. The head joint’s position was almost unchanged. We calculated the angle between hand joint and head joint and the angle between head joint and elbow joint to describe the joints’ position more accurate.The experimental results show that the proposed feature extraction method is effective and the accuracy is improved.Secondly, after the feature extraction, the method of dynamic gesture recognition is studied. We choose the KNN recognition algorithm which is based on matching thought to recognize dynamic gesture. Aiming at the lack of KNN algorithm, we proposed fast recognition KNN algorithm based on spherical region. We verified the effectiveness of theproposed method by experiments. The recognition time is reduced and the correct rate of recognition is almost invariant.Finally, we designed five dynamic gestures to satisfy the requirement of communication with home service robot. We conducted real-time recognition in different skeleton size and different light intensity. We analyzed and summarized the results of the experiment. We verified the possibility of the method proposed in this thesis to use in practical application.