Intellectual Property India Publishes Patent Application for A Real-Time Computer Vision And Deep Learning Framework For Sign Language Recognition And Natural Language Translation

MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641106657 A) filed by Indu B; Dayananda Sagar Academy Of Technology And Managemnt; Karthik Bn; Eshaa S Gurumath; Tarun Sanjay Rao; and Prachi Uday Jakati on September 04, 2026, for A Real-Time Computer Vision And Deep Learning Framework For Sign Language Recognition And Natural Language Translation.

Inventors include Indu B; Dayananda Sagar Academy Of Technology And Managemnt; Karthik Bn; Eshaa S Gurumath; Tarun Sanjay Rao; and Prachi Uday Jakati.

The application for the patent was published on September 11, 2026, under issue no. 37/2026.

Abstract: An important aspect of daily life for every human being is communication, its accessibility remains a significant challenge for individuals with hearing and speech disabilities. This work presents a novel, real time system to translate American Sign Language (ASL) word level gestures into text. A primary challenge in ASL translation is the close resemblance of gestures with other gestures, using traditional classifiers or single stage classifiers results in poor accuracy as they fail to capture dependencies over a period of time. Our system overcomes this by introducing a Two Stage Hybrid Framework. First, in stage I the system acquires video from a webcam and uses a python library named Mediapipe to extract landmarks of hands and the pose of the user to create a skeleton representation. A sequence is a series of video frames in a specific temporal window.The system then feeds the video to a Bidirectional LSTM (Bi-LSTM) model in the form of sequences, capping the number of frames to 30, the Bi-LSTM model then analyses the movement of keypoints or landmarks over a period of time and classifies it into glosses or raw words. In stage II the raw predicted sequence of glosses is passed to a T5 Transformer that understands the semantic context and excels at correcting grammatical errors.The hybrid system was developed on a standard Nvidia RTX 3050, the system achieves 99.0% training accuracy, offering a viable, low latency solution for accesible communication.

Disclaimer: Curated by HT Syndication.