MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202611002259 A) filed by Manipal University, Jaipur, Rajasthan, on Jan. 8, for 'a parallax-aware attention-driven deep image stitching method for aerial drone applications.'
Inventor(s) include Kashish Tiwari; Aman Tiwari; and Dr. Rekha Chaturvedi.
The application for the patent was published on Feb. 13, under issue no. 07/2026.
According to the abstract released by the Intellectual Property India: "The present invention relates to an AI-based aerial image stitching method for geotagged drone images. The method comprises acquiring overlapping aerial images of a target geographical area using a drone camera module, followed by a pre-processing stage including noise removal and blank frame recognition. A deep feature encoder-decoder pipeline is applied to extract multi-scale, context-aware visual features, which are subsequently matched across overlapping image pairs using a transformer-based attention alignment network. The invention employs a dual-stage geometric modelling framework, wherein a global homography is estimated using attention-enhanced correspondence filtering for coarse structural alignment, followed by locally applied Thin Plate Spline-based non-linear warping to compensate for parallax and non-planar terrain deformation. An attention-driven fusion module adaptively selects regions requiring global or local alignment. Neural seam optimization and artifact removal modules ensure perceptually consistent mosaicing. The trained model supports real-time inference on edge GPU-enabled drones. Performance evaluation using PSNR, SSIM, and dimensional alignment metrics demonstrates superior image quality compared to existing methods. The invention is applicable to agriculture mapping, disaster assessment, surveillance, and urban planning."
Disclaimer: Curated by HT Syndication.
