Survey on Advanced Computational Techniques for Sign Language Gesture Interpretation

Abstract

The interpretation of sign language motions is critical for improving communication accessibility for deaf and hard-of-hearing people. This research proposes a comprehensive computational framework for feature extraction and Long Short-Term Memory (LSTM) networks to capture temporal dynamics across gesture sequences. The CNN architecture is used to evaluate visual inputs, successfully recognizing and categorizing hand shapes, face expressions, and body postures that are critical for proper gesture interpretation. By adding LSTMs, our method effectively replicates the sequential nature of sign language, allowing for the identification of continuous gestures impacted by previous movements. We use numerous innovative strategies to handle the issues of sign language detection, such as variety in signing styles, surrounding noise, and the need for real-time processing. Multi-modal data fusion incorporates visual, contextual, and language information to improve model robustness. Rotation, scaling, and temporal shifting are used as data augmentation procedures to increase the training dataset and improve model applicability across a variety of signing settings. The hybrid CNN-LSTM architecture is enhanced via hyper parameter tuning, dropout regularization, and batch normalization to reduce overfitting while preserving excellent.