Abstract
The development of a reliable model of recognition of Kazakh sign language is an important step towards the development of inclusive communication and assistance to people with hearing impairments. This paper describes in detail the process of collecting and annotating data in which gesture images were used. Special attention is paid to the preparation and preprocessing of data to ensure their compatibility with the model. The process of learning the model involves optimizing hyperparameters and using various techniques to improve recognition accuracy. We also conducted a comprehensive performance assessment of the model based on test data to ensure its effectiveness in real-world conditions. In addition to the main development phase, we are considering testing the YOLO-NAS model on the same dataset to explore potential improvements in accuracy and performance. In conclusion, the results of our research can be used to further develop technologies that facilitate the integration of people with hearing impairments into society, as well as to create educational and communication platforms based on the Kazakh sign language.