Converting sign language into text using deep learning methods

Individuals with hearing and speech impairments experience serious difficulties in communicating with their surroundings. This thesis aims to address this issue with a real-time sign language recognition system. Within the scope of the study, a dataset consisting of 17 frequently encountered words in the literature was specially prepared by 5 subjects. The key points of the words were successfully extracted using the human pose estimation system in the MediaPipe library. These key points were classified using the LSTM architecture of deep learning models, and the obtained results provided 99% accuracy. This proposed system offers an effective solution to improve the daily communication of individuals with hearing and speech impairments.