MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101779 A) filed by Vellore Institute Of Technology on August 23, 2026, for System And Method For Sub-Pixel Motion Signature Learning And Ultra-Fine Anomaly Detection In Manufacturing Video Streams.
Inventors include Madhu Viswanatham V; Kuchipudi Vinay Kumar; Praneeth Ravi; and Romith Nirvan Edara.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: The present invention relates to a system and method for ultra-fine anomaly detection in manufacturing environments using sub-pixel motion signature learning. The system comprises a video input module (101), a video pre-processing module (102), a sub-pixel motion estimation module (103), a motion feature extraction module (104), a deep learning model module (105), an anomaly detection module (106), and an output and alerts module (107). Video streams acquired from industrial processes are pre-processed and analyzed using optical-flow-based sub-pixel motion estimation to detect minute motion variations. Motion features are extracted and processed by deep learning models to learn normal motion signatures and temporal behavior patterns. Real-time motion data is compared against the learned signatures to identify deviations indicative of equipment degradation, process abnormalities, or manufacturing defects. Detected anomalies are scored and communicated through visualizations and alerts, enabling predictive maintenance, reduced downtime, improved operational reliability, and enhanced product quality.
Disclaimer: Curated by HT Syndication.
