The System Implementation And Research Of Static Digital Sign Language Recognition Based On Kinect

With the rapid development of science and technology, the interaction between man and computer is developing from the initial command line interface, graphical user interface towards more intelligent natural user interface.Users interact with the computer in a natural way such as gestures, voice,touch,is called natural user interface. Kinect is a motion-sensing camera launched by Microsoft, it can real-time capture the depth image, color image, voice and other information of the human body.In this paper, by using kinect for single-handed static sign digital language recognition, providing a non-contact human-computer interaction. The research work of this article mainly includes the following several aspects.Firstly, this paper proposes a binary image denoising method of static digital sign language by looking for the biggest connected area. The method can remove the noise of the binary image. For feature extraction of image, with Height Function algorithm the recognition rate increased by 2.38% compared to that before denoising. For the Inner-Distance Shape Context algorithm the recognition rate increased by 1.68% compared to that before denoising.Secondly, by using Height Function and Inner-Distance Shape Context to extract image features. With both algorithms for the extraction of feature vector for denoising before and after of the binary image, it was found that the former has higher recognition rate.Thirdly, sign language binary image classification based on support vector machine(SVM). Using support vector machine(SVM) to extract the image feature vector for double cross validation experiments 10 times. For the same kind of feature extraction algorithm, this paper compares and analyzes the image recognition rate for denoising before and after. The conclusion is that for binary image after denoising and feature extraction with Height Function, image recognition rate is highest.Fourthly, the establishment of an image database which contains 500 binary images. We recruited 10 volunteers to collect images. They collect images according to the order of 0 to 9, each sign language acquisition 5 images according to different situations.Fifthly, the design and development of a static digital sign language recognition system. The system decides whether to start reading sign language images through the comparison of the vertical ordinate value between hand node and elbow node. The system began to image denoising and feature extraction, and finally bringing feature vector to the trained SVM model, then output results.