Edge-Optimized AI-Powered Translator for Indian Sign Language (ISL)

Sign Language translation bridges the communi- cation gap between the Deaf and the hearing communities. According to the World Health Organization over 430 million people worldwide experience hearing loss, including nearly 18 million Deaf individuals in India who rely on Indian Sign Language(ISL). However, continuous sign language translation presents significant challenges due to complex spatio-temporal dependencies, signer variations, and contextual ambiguity. This work proposes an efficient, edge-based deep learning framework for continuous ISL translation using landmark-based motion representations. By extracting structured hand, pose, and facial keypoints instead of raw RGB frames, the system reduces com- putational complexity while preserving linguistic information. An ensemble-based classifier with probabilistic modeling ensures robust recognition, while a lightweight student model enables mobile deployment. Experimental evaluation on the dataset demonstrates reliable recognition performance with low-latency inference, validating the suitability of the approach for real-time, resource-constrained environments.