Рақамли технологияларнинг назарий ва амалий масалалари Ҷилди 8 № 3 (2025) · Саҳифаҳои 82-93
Technology for Preparing a Dataset Based on Mediapipe for Translating Text to Uzbek Sign Language
Джураев, Д.Б.
Аннотатсия
This paper discusses the problem of creating a multimodal dataset based on MediaPipe technology for automatic translation of Uzbek sign language into text and speech. Due to the multifaceted nature of sign language hand movements, facial expressions, body posture, and gaze direction the technical, linguistic, and ethical aspects of dataset preparation are analyzed. Using MediaPipe, key points of the hands, face, and body are extracted, and experiments are conducted with GRU, LSTM, and BiLSTM models. Comparative analysis demonstrates the efficiency and resource-saving capability of the GRU model, while LSTM and BiLSTM show advantages in processing complex sequences. Thus, the multimodal dataset developed in this study provides a reliable foundation for the development of real-time sign language recognition systems.
жестовый языкMediaPipeмультимодальный датасетRNN (GRU, LSTM, BiLSTM)компьютерное зрениеобработка естественного языка (NLP)искусственный интеллектsign languageMediaPipemultimodal dataset
Манбаи метамаълумот: бойгонии OAI-PMH-и маҷалла · Sindex матни пурраро нигоҳ намедорад, ба манбаъ пайванд медиҳад.