Real-Time Sign Language Recognition and Communication: Leveraging CNN for Deaf and Mute Communities

Accessibility, communication, and empowerment issues are common for people with sensory disabilities. Although this problem has been the subject of many studies, there is no one-stop application. The accessibility platform is a platform-independent application accessible through a website. This initiative seeks to assist in resolving these issues through the real-time translation of Indian sign language into text. Additionally, it would provide users the opportunity to connect with NGOs that focus on empowering people with disabilities. It is web-based making it platform-independent. The system aims to create a comprehensive, all-in-one feature platform, rather than focusing on individual features. The Indian Sign Language feature uses a Convolutional Neural Network as the algorithm for image classification, it uses Leaky ReLu and Adams optimizers. The use of these optimizers ensures that the accuracy of the research reaches above 90 percent in comparison to the accuracy of other networks that use sigmoid and tanh activation functions. The system achieved an accuracy of 97 percent in real time with the data set of 1200 images and 35 classes i.e. 26 alphabets and 9 numbers.