MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202611066078 A) filed by Manipal University Jaipur on May 26, 2026, for A System And Method For Real-Time Privacy-Preserving Sign Language Recognition On Edge Hardware Employing Landmark Canonicalization And Hybrid Temporal Classification.
Inventors include Krish Ray; and Dr. Mayank Namdev.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The present invention relates to a system and method for real-time sign language gesture recognition on resource-constrained edge hardware. The method comprises extracting normalised 3D hand landmarks on-device without transmitting raw image data, canonicalization to achieve invariance to translation, scale, rotation, and handedness on-device. Temporal buffering is used to form rolling landmark windows, a deterministic geometric rule engine, and a temporal sequence model. Output stability is enhanced using exponential moving average smoothing, consecutive-frame confirmation, and velocity-based gating to reduce spurious predictions. The system operates fully on-device without transmitting or storing raw image or video data, transmitting only anonymized canonical landmark tensors for low-confidence logging. It achieves a 0.27 MB model size, 2.37 ms inference latency on ARM hardware, over 80% Top-1 accuracy, and 83.81% Top-3 accuracy, suitable for privacy-preserving real-time assistive applications.
Disclaimer: Curated by HT Syndication.
