Technology Of The Analysis And Generating Of Sign Language Data

Sign language is a static expressing system that is composed of signs by using hand motion aided by face expressions. Sign Language Recognition (SLR) can help the deaf and provides an effective and correct mechanism to translate sign language into texts or common language, to make it more convenient to communicate between the deaf and the normal by computers. SLR based on data gloves may offer natural human computer interactions. Thus SLR has tremendous research necessity.In the research of person-independent sign language recognition, the contradiction brought about by the difference among data and the lack of training samples has made the sign language data an imperative problem to be analyzed. We did the following work to resolve this problem:1. A new framework based on generating data driving for SLR is proposed in this paper. In this framework, valid sign language generating algorithms can be used to generate new samples from the original data set, which would be used to train the model, and solve the lack of the training samples successfully. We introduce two sign language generating algorithms, based on Genetic Algorithm and Mean-shift respectively, and inspect the impact on the performance of the system by employing the two synthesis algorithms in the new framework. The experimental result shows that the recognition rate improves evidently by adopting the mean of generating data driving on the unregistered test set. The recognition rate reaches to 67.3% by using the generating data driving based on Genetic Algorithm, and the recognition rate reaches to 71.5% by using the inward generating data driving based on Mean-shift on the optimized generating parameters, and increases the recognition rate by 5.1%compared with original train data driving.2. The structure and variation of sign language is analyzed based the understanding of sign signal from the perspectives of human kinemics and linguistics. Our method extracts the key actions embodying the commonness of sign language in signs to ensure the structurality of sign signals, describes and defines the basic movement features correlated with the individuality of signs to build the sign language expression model, including movement trajectory, timing, and hand shape.3. Based on the existed Chinese sign language synthesis system, the expression model is employed to generate a great deal of valid and general data, which would be used to enlarge sign language database, and drive the virtual person to display sign language full of individuality.