Study Of The Sign Language Recognition System Based On Monocular Vision

Sign language recognition, which is based on monocular vision, is a natural and intuitive interaction.It is collected by the camera to identify the image recognition,without additional any auxiliary equipment. This is a significant research topic, will to promote the development of computer vision. The System can help improve the living, learning and working environment for the deaf-dumb people, and creating a more natural human-computer interaction environment.In this paper, on the basis of sign language recognition research at home and abroad, studies image preprocessing, gesture detection, gesture tracking, gesture feature extraction and sign language recognition. Specific research work includes the following aspects:(1)For Hsu R.L light compensation algorithm easy to take the skin color as the "reference white" in the condition of the dark light, Increase compensation judgment conditions in the algorithm and reduce the intensity of compensation to improve the skin color segmentation effect, and the effectiveness of the method is verified by experiments.(2)Fusion of Skin color detection and binarization algorithm, in order to reduce the number of times to traverse the image and the time of reading and writing pixels. Put forward a gesture detection method based on YCgCr and YCgCb skin color model, And through experiment to verify the method of the gestures localization effect in different light and environment.(3)In the detection of both hands, put forward a new way for both hands detection based on the structure of human. The way is according to the human face to divided the image into four candidate area. Exclude A and B areas, detected of the right hand and left hand in C and D areas, by narrowing the image detection area to improve the execution efficiency of the algorithm.(4)For the CamShift target tracking error and targeting loss, put forward a way of gesture tracking based on skin color model and CamShift. The way can track the target by the centroid of the hand gesture contour and the centroid of the Cam Shift track window,And the experiment results verify the tracking effect. In addition, gives a realization method and process of the both hands tracking.(5)Studied of the sign language recognition based on SURF-BoW feature. Builded the image database of Chinese finger language. Implemented a sign language recognition system, which can collecting image, gesture segmentation, hand tracking and recognizing on real-time.