Recognition of Signs and Movement Epentheses in Russian Sign Language

Automated translation from sign languages used by the hearing-impaired people worldwide is an important but so far unresolved task ensuring universal communication in the society. In our paper we propose an original approach towards recognition of Russian Sign Language (RSL) based on extraction of components: handshape and palm orientation, location, path and local movement, as well as non-manual component. We detail the development of the dataset for subsequent training of the artificial neural network (ANN) that we construct for the recognition. We further consider two approaches towards continuous sign language recognition, which are based on sequential search of candidate events for the next sign start and the complete identification of the speech elements - the actual signs, resting state of the signer, combinatorial changes in the parameters of the signs and the epentheses.