LSTM based Pose Estimation and Sign Language Translation

A vital communication tool for those who are hard of hearing or deaf is sign language. However, communication is severely hampered by the general public's poor comprehension of sign language. There different sign languages software for helping the deaf and dumb people. In order to use those, there is a to need specified software with high end or medium end hardware devices. In order to run-in low-end hardware devices this problem is addressed, a real- time system for translating Indian Sign Language (ISL) that uses deep learning and position estimation to enable fluid interpretation of hand and body movements with limited actions was suggested. Using a complex posture estimation architecture called MediaPipe Holistic, this method recognizes important skeletal landmarks such as torso, face, and hand points. An RNN model for deep learning that is especially made for processing sequential data, the Long Short-Term Memory (LSTM) network, is used to assess these properties After training and testing the proposed model the accuracy of 95% is obtained for recognizing the actions in low wend hardware like without GPU. It shows how efficiency and friendly use of all low-end hardware devices that works on only CPU.