MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202621065308 A) filed by Kedar Sanjay Gurav; Prof. Deepa Abin; Abhimanyu Balasaheb Gitte; Harsh Abhishek Gupta; Aditya Suresh Gupta; and Vishwakarma Institute Of Technology on May 23, 2026, for A System And Method For Monocular Depth Estimation For Industrial Topological Inspection.
Inventors include Kedar Sanjay Gurav; Prof. Deepa Abin; Abhimanyu Balasaheb Gitte; Harsh Abhishek Gupta; and Aditya Suresh Gupta.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The present invention provides a hardware-constrained monocular depth estimation system and method for industrial topological surface inspection. The system employs a pseudo-label knowledge distillation framework in which one or more large-capacity pre- trained depth teacher models generate geometrically filtered per-pixel depth supervision targets for training a compact student depth estimation network deployable on resource-constrained edge-computing hardware. A composite training loss integrates a scale- invariant depth regression term, an edge-guided gradient matching term that selectively enforces depth discontinuity alignment at geometrically meaningful image boundaries whilst suppressing penalty at texture-only transitions, and a surface-normal consistency term derived from cross-product normal vector computations on the predicted depth map. An explainability validation module generates gradient-weighted spatial activation attribution maps from the student network enabling real-time classification of depth predictions as geometry-driven or texture-driven, providing quantitative prediction confidence scores for industrial inspection reliability assessment. An optimised inference pipeline incorporating model quantisation, operator fusion, and adaptive memory management enables deployment within the processing and memory constraints of edge-computing hardware. The system produces per-pixel depth maps, surface normal maps, and explainability-annotated inspection reports suitable for automated defect localisation and topological quality assessment in industrial manufacturing and infrastructure inspection applications.
Disclaimer: Curated by HT Syndication.
