Neural Network-based Real-Time Recognition of American Sign Language Finger-Spelled Gestures: Bridging Communication Gaps
Soumya Ashwath, Ashwin Shenoy M
Abstract
In this work, an innovative solution addresses the limitations of sign language communication, arising from a lack of familiarity and interpreter availability, by leveraging advanced neural networks for real-time recognition of finger-spelled gestures in American Sign Language. This approach bridges the communication gap by providing a means to decode sign language gestures, regardless of the user's familiarity with sign language, through a two-step process involving gesture refinement and classification. The ultimate objective is to create a model capable of seamlessly translating live video feeds of American Sign Language hand gestures into on-screen text representations of specific signs. Notably, our meticulously trained CNN classifierachieves an impressive 95.7 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> accuracy in recognizing 27 symbols, including the 26 English alphabets and a ‘blank’ symbol usedfor sentence spacing.