MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124573 A) filed by Pokkunuri Pardha Saradhi; and Koneru Lakshmaiah Education Foundation on October 17, 2026, for Physics-Plausibility-Gated Kinetic-Coupling Graph Network With Evidential Multi-Domain Fusion For Robust Skeleton-Based Human Action Recognition..
Inventors include Supriya Pacha; and Pavan Kumar T.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: The present invention relates to a system and method for robust skeleton-based human action recognition. A differentiable kinematic front-end computes smoothed velocities and accelerations and, using learnable anthropometric masses, the kinetic energy and momentum of every joint. A physical-plausibility estimator scores each joint in each frame from biomechanical tests. These scores gate message passing in a kinetic-coupling dynamic graph built for each temporal window, weight physics- based joint–object contact links, and discount the evidence of the joint, bone, kinetic and spectral streams before Dempster–Shafer fusion. Spectral-entropy routing lets each joint select its temporal receptive field. The system outputs an action label with a calibrated uncertainty and, when the uncertainty is high, withholds the decision and raises a sensor-fault alert. The invention therefore contributes to reliable, physics-aware and uncertainty-aware human action recognition for monitoring, safety and human–robot interaction.
Disclaimer: Curated by HT Syndication.
