MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202521134046 A) filed by Ashwini Nikhil Shinde; and Dr. Rajeswari Kannan, Pune, Maharashtra, on Dec. 31, 2025, for 'a video sequence learning system and a method for gesture-to-text translation.'
Inventor(s) include Ashwini Nikhil Shinde; and Dr. Rajeswari Kannan.
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 a video sequence learning system and method for gesture-to-text translation is disclosed. the system receives sign-language videos, pre-processes each video by extracting frames, cropping hand-and-gesture regions, and applying illumination-preserving, high-contrast enhancement. an augmentation module generates multiple variations per video using techniques including noise injection, blur, clarity reduction, flipping, rotation, resizing, temporal speed adjustment, and brightness/contrast modifications, compiling outputs into a shuffled dataset. a two-pass feature-extraction pipeline computes frame-level Histogram of Oriented Gradients features, zero-pads vectors to a maximum length, aggregates to video-level representations, and applies principal component analysis for fixed-length embeddings. sequence learning employs an encoder decoder recurrent neural network with attention, batch sorting by sequence length, scheduled teacher forcing, and parameter initialization from Glorot, orthogonal, Gaussian, or positive forget-gate bias schemes. the system and method evaluate model performance using accuracy, BLEU-N, ROUGE-L, and METEOR metrics, exporting structured outputs, and achieving improved translation performance relative to baseline attention models."
Disclaimer: Curated by HT Syndication.
