Dynamic Gesture Recognition Based On Kinect Depth Data

With the rapid development of science and technology, the quality and level of the human-computer interaction improves Continuously. As an alternative to traditional input devices (keyboard,mousejoystick,etc.), gestures have been as an important way of the pursuit of ideal natural interaction, their study also has important theoretical research significance and practical application value.Based on the current research of hand gesture recognition results, the thesis analyzes the advantages and disadvantages of various methods.In order to realize the human-computer interaction better, this thesis designs a kind of hand segmentation,tracking,recognition and recognition system through the depth image obtained from the device camera.Accurate gesture is the first condition of gesture recognition, because in the process of processing the depth image, there will be the problem of false segmentation for hand wrist. And it is vulnerable to the similar color object and easily disturbed by light factors when dealing with color images. Therefore.this thesis firstly uses the camera provided by Microsoft to obtain depth image. And it eachly uses the method of double threshold segmentation and Gaussian skin color model to process the depth images and color images obtained from Kinect sensor. Finally it realizes the segmentation and recognition of hand gesture.This method can effectively eliminate the influence of adverse factors such as illumination and background on the result of gesture recognition. A large number of experiments shows that the proposed method is accurate, effective and it has good robustness. After hand gestures segmentation, the next step is to track hand gesture.this thesis combines CamShift algorithm and Kalman filtering to track hand-gesture. Gesture feature extraction is the crucial step in the gesture recognition.This thesis uses the hidden markov model (HMM) to train the extracting characteristics.The thesis adopts the Baum Welch algorithm as the training method.The computation efficiency is greatly increased.And this thesis uses the threshold of the HMM model to distinguish the undefined gestures. This thesis designed five gesture recognitions of front, left, right, up, down. And it verified the experimental results through the practical application. The experimental results show that tracking results are accurate and robust.