Russian Sign Language is a natural language that serves as a means of communication for people with hearing impairments. Currently, it is necessary to provide communication paths between hearing and deaf people, which requires the solution of several specific problems, one of which is gesture recognition. The paper presents a system for recognizing isolated static and dynamic gestures of the dactyl alphabet of the Russian Sign Language. The system is based on machine learning methods and is an LSTM with Mediapipe Hands as feature extractor. The network showed an F-measure value of 91% on test data.