Enhanced Indian Sign Language Detection

Abstract - The communication problem involving members of society who have speech and hearing impairments is still not fully resolved. In an earlier study, we created a real-time Indian Sign Language (ISL) recognition system which uses LSTM architecture for sequential gesture recognition. The focus of this paper is on further improving this system by changing the architecture from LSTM to CNN to enhance spatial feature extraction and overall system performance. Using a more comprehensive ISL dataset, we trained and tested the model and added new advanced preprocessing techniques such as Gaussian blur and converting the images to grayscale. These modifications improved the accuracy of the model and reduced the processing power needed, allowing for more advanced, rapid, and reliable real-time ISL gesture recognition. The result of this study is in the direction of making available an effective, simple, and easy-to-use technological interface for the deaf and hearing-impaired people in India. Keywords: Indian Sign Language (ISL), Gesture Recognition, Convolutional Neural Networks (CNNs), Real-time Communication, Image Preprocessing, Gaussian Blur, Grayscale Conversion, Sign Language Translation, Computer Vision, Human-Computer Interaction (HCI), Deep Learning