MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621074084 A) filed by Ashokrao Mane Group Of Institutions, An Autonomous Institute, Kolhapur on June 15, 2026, for Unsupervised Human Activity Recognition In Industrial Disassembly Using Deep Clustering And Temporal Analysis.
Inventors include Prof. Anil B. Desai; Prof. Suraj S. Redekar; Diksha A. Jadhav; Prof. Satish P. Pise; Prof. Sarita A. Chavan; and Mr. Yashvardhan Yuvraj Patil.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention relates to a method and system for unsupervised human activity recognition specifically designed for industrial disassembly tasks. The invention addresses the critical limitation of existing supervised approaches that require expensive manual annotation by introducing a novel unsupervised learning framework comprising deep embedded clustering and temporal pattern analysis. The system captures video data of disassembly operations, extracts spatio-temporal features using a pre-trained 3D-CNN backbone, and learns unsupervised representations through a deep embedded clustering network that jointly optimizes reconstruction and clustering losses. A temporal convolutional network with dilated causal convolutions models sequential dependencies among disassembly activities, enabling automatic discovery of task sequences and procedural constraints. An integrated anomaly detection module identifies deviations from learned patterns for real-time safety monitoring. The invention eliminates annotation costs, automatically discovers activity patterns, adapts to new environments without re-annotation, and supports scalable deployment in manufacturing, e-waste recycling, and maintenance operations.
Disclaimer: Curated by HT Syndication.
