Dynamic gesture recognition using hand pose-based neural networks for sign language interpretation

The dynamic two-hand Sign Language Recognition system is effective for individuals with speech and hearing impairments, and it also plays a crucial role in computer vision research. Additionally, real-time recognition of such gestures presents a significant challenge as the machine must accurately identify combinations of patterns and interpret them within specified time constraints. The current vision-based systems designed for dynamic gesture recognition are deemed ineffective as a result of an excessive amount of feature extraction from lengthy sequences of frames. Furthermore, current systems are exclusively designed for real-time static gesture recognition. Therefore, this paper proposes a vision-based dynamic sign language recognition system using Time Distributed Long Short-Term Memory Networks that addresses the deficiencies of current systems. The pipeline for creating the corpus for the proposed Dynamic Two Hands Indian Sign Language dataset is also worked out. The proposed model utilizes a keypoints-based feature extraction technique to enhance the accuracy and efficiency of the system. The experimental findings indicate that the developed system has a recognition rate of 99.6%surpassing other current state-of-the-art methods.