Dactylology Interpretation Using Key-Points Detection and LSTM

The practice of finger spelling, or dactylology, is frequently used by deaf, hard of hearing, mute, or people working underwater. It is also used when speaking out loud is not an option or practical. In this paper, key-point detection and Long Short-Term Memory (LSTM) model are used to demonstrate a method for deciphering dactylology. A LSTM network trained to recognize and interpret finger spelling gestures is fed with key-points from a real-time video sequence to detect the faces, postures, and hands. On the basis of annotated dactylology video sequences, this method is evaluated for its accuracy in interpreting a wide variety of finger spelling gestures. In real-time, the key-point detection method with the LSTM approach shows promise for interpreting dactylology. By using key-point detection and LSTM networks, finger spelling can be made efficient and accurate through key-point detection and LSTM networks. Many applications are possible with this approach, such as translating sign language and developing assistive technology for the deaf.