DrishtiSign: An Intelligent Sign Language Interpreter for Indian Languages

Sign language remains one of the most important yet underserved areas of accessible communication.Despite its significance, most learners struggle to find tools that go beyond static tutorials or scheduled sessions with an instructor.This paper presents DrishtiSign, an AI-powered, camera-based hand sign recognition and interactive tutoring system that uses a standard webcam to recognize ASL (American Sign Language) and ISL (Indian Sign Language) hand signs in real time and guide learners toward accuracy.The system extracts 21 threedimensional hand landmarks per frame with the MediaPipe Hand Landmarker pipeline, normalizes them relative to the wrist joint, and classifies the resulting 63-float feature vector with a deep neural network that has a residual skip connection.Temporal smoothing by majority voting over a 15-frame rolling window keeps the predictions stable.An interactive tutoring module asks users to perform specific alphabets and moves ahead only after a correct, confident match.Experimental results show that the proposed GestureClassifier reaches 97.983% accuracy across the ASL alphabet, eight additional whole-word signs (Bye, Deaf, Hello, Not Ok, Pen, Please, Thank you, Yes) and three control classes (space, del, nothing), for 37 classes in total, and handles an Indian Sign Language consonant set of 37 classes (BA, CHA, DHA, KHA, GA, MA, and others) with a held-out test accuracy of 80.1%.The system needs no specialized hardware and runs in real time on standard CPU-based devices.