Sign Language Recognition using BiLSTM model
S Renjith, Rashmi Manazhy, Mr. S. Suresh
Abstract
Sign Language Recognition (SLR) is a crucial tool that facilitates efficient communication for people with hearing disabilities. This research investigates the use of Bidirectional Long Short-Term Memory (BiLSTM) neural networks to recognize sign language. Two distinct datasets, namely the German Sign Language dataset and the MNIST Sign Language dataset, are employed. Both datasets consist of 26 classes, each representing different alphabets. The goal is to develop a dependable and accurate model for comprehending the movements of sign language. The BiLSTM design’s capacity to retain sequential information from both the past and future makes it well-suited for the dynamic and temporal characteristics of sign language. The MNIST collection incorporates signs from American Sign Language (ASL), whereas the German Sign Language dataset is used to recognize German finger-spelled alphabets. These datasets provide a comprehensive assessment of the model’s capabilities due to the variety of signals and movements they include. The BiLSTM model exhibited impressive accuracy rates of 92.2 percent and 96.6 percent on the German and MNIST datasets, respectively, as seen by the experimental findings. This work contributes to the development of advanced sign language detecting systems, which have the capacity to enhance accessibility and communication for individuals with hearing impairments. The findings illustrate the versatility of BiLSTM models in sign language recognition and their potential use in real-world scenarios.