Gesture Recognition Based On Kinect

With the continual breakthrough and development of science and technology,the interaction between human and machine becomes more and more common.Gesture as a natural and intuitive interaction mode,with a strong visual impact.In recent years,gesture recognition has become one of the hot research areas.Based on the visual gesture recognition,it is the key technology indispensable for the next generation of human-computer interaction.June 2010 Microsoft has introduced Kinect,a somatosensory peripheral that captures depth information and enables tracking and recognition.As a revolutionary product,Kinect not only provides a new way for interaction,but also provides new technical means for gesture recognition.Gesture recognition system is divided into gesture detection and gesture recognition two processes,this paper mainly studies the two processes,and proposes the improved method for the existing problems of the existing algorithms.The main work and research contents of this paper are as follows:(1)Based on Kinect gesture recognition system to build.Acquisition of the original image data,to the depth of the image to be denoised.(2)In the research of gesture detection and extraction,the traditional Adaboost algorithm may be over-trained in the process of training,and an improved method of weight threshold judgment is proposed to segment the gesture area accurately.(3)In the research of gesture recognition,aiming at the problem of low recognition accuracy of K-curvature algorithm in traditional finger-point detection,a K-curvature finger point improvement algorithm based on convex hull is proposed to realize a class of gesture recognition.In addition,this paper also proposed a morphological finger detection algorithm,which mainly used the method of maximum inscribed circle to detect the number of fingers rapidly.(4)To carry out experimental verification.Using the improved Adaboost algorithm proposed in this paper,the detection of the gesture area is tested,and the finger detection is performed by using the improved K-curvature algorithm.Experiments show that the improved method has better robustness and higher precision.