Gesture Recognition Based On Kinect Depth Data

With the development of computer technology, human-computer interaction hasbeen moving towards sustainable development of humanity and simplistic wayconstantly, and gesture recognition is an important way of human-computerinteraction. Compared to the mouse，keyboard and other contact mode of operation,natural and intuitive gestures are more natural human behavior. Hand gesturerecognition based on computer vision high requirements on environmentalbackground, depth image obtained by the Kinect sensor can solve this problem. So theresearch of gesture recognition based on depth information to improve the interactiveeffect based on human-computer interaction, promoting technology plays animportant role in all walks of life application. This paper carried out a thorough studyon hand gesture recognition based on depth data, and developed a gesture recognitionsystem, research contents are as follows:Firstly, analyses the present research situation of the background and significanceof the study，gesture recognition at home and abroad, and gives some gesturerecognition algorithms.Secondly, research the internal structure and working principle of Kinect sensor,the specific method for pretreatment and the use of the OpenNI framework andobtaining the depth information.Thirdly, study on static hand gesture recognition algorithm based on Kinectdepth information. First the paper introduced the hand extraction algorithm. Then thefinger detection method, the traditional at the same time, put forward a threshold pointdetection method for static hand gesture, the method can accurately provide importantinformation to hand: the accurate position of finger, finger, palm, provide informationto let the hand and do real-time interactive system. The system can determine thefingers and the palm of the hand accurately under complex background position, andallow the hand and arm to a certain degree of turnover. Finally, on the base ofidentification of the finger, we realized a finger name and digital gesture recognition.Finally, make a thorough research on gesture recognition dynamic. First, we usedKinect SDK to obtain the skeletal features of user index and the human body, and thesmoothing. Then we study the principle of DTW dynamic programming algorithm,DTW algorithm according to the weight of the gesture recognition of different jointpoint is different, we proposed the weighted DTW dynamic programming algorithm, experiments show that the algorithm can effectively trained hand by statistical method,and then achieve the recognition function by skeletal joint trace characteristics andweighted analysis hand rail sequence matching, can identify waved, mobile,promoting action. The method is identified stable, and not influenced by background,distance etc., can meet the need of dynamic hand gesture recognition of remote.