Research On Vision-based Finger Language Recognition System

Deaf-mute communicates with others through sign language.Traditional sign language requires human translation which leads low efficiency.With the continuous development of science and technology,sign language recognition has become focus of research in the field of computer vision.Based on human-computer interaction platform,sign language recognition improves some shortcomings of the traditional sign language and helps the deaf-mute to express.With broad application prospects,it's importantly significant to study sign language.In the field of sign language recognition,there are two kinds of sign language recognition technology research methods which are respectively based on data glove and the vision.With the rapid development of computer interactive technology,sign language recognition technology based on visual research is more in line with historical trend.This thesis is mainly based on OpenCV open source library.The thesis presents an sign language recognition algorithm by gesture tracking,image preprocessing and SVM recognition.The main research is as follows:(1)A gesture segmentation algorithm is proposed.The difference among interframe difference method,three-frame difference method and background subtraction method is compared.The advantages and disadvantages of the algorithms are analyzed.Modeled in HSV color space,the codebook is used to segment the gesture area.The proposed algorithm is proved to be efficient.(2)The shortages of traditional Meanshift tracking algorithm and Camshift tracking algorithm are separately analyzed.The window of Meanshift tracking algorithm can't be adapted.Camshift tracking algorithm lacks of prediction mechanism,which easily leads to the target missing when blocked by interference.After experimental analysis,Kalman filter is combined with Camshift so that gesture motion can be predicted.In this thesis,this algorithm can meet the practical requirements.(3)To obtain the required the sign images,the area of hand gesture segmentation needs median filtering,morphological processing and Ostu threshold processing.Finally,the method of 8-connected neighborhood is adopted to extract continuous gestures contours.(4)A method for extracting gesture characteristics is proposed.Hu moment invariant features are chose as global features and fourier descriptor characteristics as local features of gesture contours.The combination of two characteristics is the gesture feature vector.This method is discriminating when used for gesture recognition.(5)The sign language classification and recognition methods are determined.An algorithm based on support vector machine is proposed to classify the gesture characteristics.With the analysis of the principle of SVM and considering core function and related parameters,the sign language recognition based on SVM is determined.The experiments show that support vector machine classification and recognition method in small-sample circumstances has really nice effect.The recognition rate is up to 90% which meets real-time requirements.