REAL-TIME AMERICAN SIGN LANGUAGE RECOGNITION AND SUBTITLE GENERATION USING DEEP LEARNING

Abstract

This research introduces a real-time American Sign Language (ASL) recognition and subtitle generation system leveraging deep learning and computer vision. The system captures live video, detects hand gestures, and translates them into text subtitles displayed on-screen. A Convolutional Neural Network (CNN) trained on an extensive ASL dataset forms the core of the recognition engine. The preprocessing pipeline integrates Gaussian blur, morphological filtering, and HSV color segmentation for precise hand isolation under diverse illumination conditions. The results demonstrate an overall accuracy of 96.8%, low latency, and strong generalization, making the proposed approach a viable assistive communication tool for bridging the gap between signers and non-signers.