MUMBAI, India, May 29 -- Intellectual Property India has published a patent application (202641063594 A) filed by Coimbatore Institue Of Technology, Coimbatore, Tamil Nadu, on May 20, for 'real-time indian sign language translation system using landmark-based gesture recognition and neural network classification.'

Inventor(s) include Dr. Valliappan Raman; Dr. N. Ramkumar; Preethiev Raj I; Sudharshan M Prabhu; Sujai S; and Sujit S.

The application for the patent was published on May 29, under issue no. 22/2026.

According to the abstract released by the Intellectual Property India: "The present invention is a camera-based, edge-deployed real-time Indian Sign Language (ISL) recognition and sentence construction system that eliminates the communication barrier faced by approximately 63 million hearing- and speech-impaired individuals in India. The system comprises a Raspberry Pi Camera Board for video capture, Google MediaPipe Hands for 21-point three-dimensional hand landmark extraction, a novel dual-hand slot assignment module for consistent multi-hand feature vector construction, a Multi-Layer Perceptron (MLP) classifier supporting 66 ISL gesture classes with 94.7% accuracy, a majority-vote temporal smoothing buffer, and a novel hold-to-confirm interaction mechanism for intentional sentence construction. The entire pipeline executes at 32 frames per second on a Raspberry Pi 5 single-board computer without any GPU, internet connection, wearable hardware, or external computing device. The system is accessible from any web browser on the local network via a Streamlit web application featuring a flicker-free background-threaded camera architecture. The complete hardware cost is approximately Rs. 51,116 (inclusive of GST), representing a 60-90% cost reduction compared to GPU-based alternatives. A software simulation mode enables gesture testing using pre-recorded landmark data without a live camera. The invention is extensible to a larger ISL vocabulary, additional Indian languages, and future mobile deployment via TensorFlow Lite."

Disclaimer: Curated by HT Syndication.