Recognition of Russian and Indian Sign Languages Based on Machine Learning

In the paper, we consider recognition of sign languages (SL) with a particular focus on Russian and Indian SLs. The proposed system includes five components: configuration, orientation, localization, movement non-manual markers. analysis uses methods individual gestures continuous speech for (RSL). To recognize gestures, RSL Dataset was developed, which more than 35,000 files over 1000 signs. Each performed 5 repetitions at least by deaf native speakers Sign Language from Siberia. isolate epenthesis RSL, 312 sentences were selected recorded video. Five types movements distinguished, namely, "No gesture", "There is "Initial movement", "Transitional "Final movement". markup highlighting carried out Supervisely.ly platform. A recurrent network architecture (LSTM) built, implemented using TensorFlow Keras machine learning library. accuracy correct 95 %. work similar dataset both language (ISL) continuing. hand mediapipe holistic library module used. It contains group trained neural algorithms that allow obtaining coordinates key points body, palms face person in image. 85 % achieved verification data. future, it necessary to significantly increase amount labeled components, number rules have been developed certain face. These include positions eyes, eyelids, mouth, tongue, head tilt.