Long Short Term Memory Based Sign Language Detection System

Sign language recognition plays a crucial role in facilitating communication and inclusivity for individuals with hearing impairments. This research paper proposed a novel approach for detecting sign language gestures using Long Short-Term Memory (LSTM) networks. By leveraging the sequential data processing capabilities of LSTM networks and with the use of feature engineering an accurate model has been developed to predict sign language gestures. The paper extends in-depth discussions into data collection, data pre-processing and feature engineering techniques used to increase the efficiency of the LSTM model. Keras API for TensorFlow was used for creating a sequential model. The paper also presents a comparative study regarding the change in accuracy resulting from the change in the size of the LSTM layer and the dropout layer ratio. The highest-performing model with an accuracy of 91.28 percent was used for testing the performance of the model in real-life applications.