Sign language recognition using improved 3D convolutional neural networks

Sign language recognition (SLR) plays an important role in enabling communication for those who are hard to hear or deaf. SLR involves recognising and translating signs into natural language, and this task can be enhanced by employing deep learning methods. The proposed approach uses 3D convolutional neural networks (3D CNNs) to extract features. Through this method, improvements in the accuracy and real-time performance of SLR can be achieved. In this experimental study, a 3D CNN along with an LSTM architecture is implemented for feature extraction in SLR systems, and their advantages and limitations over 3D CNN and 2D CNN and 3D CNN combined models are highlighted. Compared with the traditional 3D CNN architecture, the 3D CNN-LSTM model can effectively interpret the spatiotemporal features of sign language expression, which is crucial for accurately recognising signs. Additionally, various strategies for optimising the architecture of 3D CNN-LSTM to achieve better performance are discussed in this article. Finally, some remaining challenges and future research directions in this area are highlighted. The analysis of the outcomes indicates that the 3D CNN-LSTM architecture has excellent potential for enhancing the accessibility of SLR systems and facilitating communication for individuals who communicate via sign language.