Computer Science · Human-Computer Interaction
A CNN Based Sign Language Learning System For Deaf & Mute User
Maisha Hasnin, Iffat Ara Afrose, Sazzadur Rahman Himel, Sayma Alam Suha, Muhammad Nazrul Islam
See on the mapAbstract
Sign language has emerged as a crucial medium of communication for individuals with hearing impairments. To effectively engage with this community, it is imperative to learn sign language. Thus, the objective of this research is to design and develop an intuitive sign language learning system based on machine learning approaches. To attain this objective, a usable and intuitive application is developed adopting the Convolutional Neural Networks (CNN) for accurate recognition of American Sign Language (ASL) gestures. The proposed CNN model is trained on a large dataset of ASL gestures to accurately classify each gesture. In addition, this research explores the integration of other models to improve the accuracy of the classification results. The study showed that the the accuracy of the proposed ASL classification model is 99.89 % on the test set, which indicates that the proposed model is suitable for reliably identifying and categorizing sign language motions. An intuitive and usable application is developed with the integration of the proposed CNN model that takes real-time input from a webcam and converts the gesture into text output. The evaluation study conducted through the System Usability Scale (SUS) approach confirms the usability and effectiveness of the system for learning and communicating using sign language. Thus, this research has the potential to significantly advance the recognition and transmission of sign language using machine learning,