MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202611074312 A) filed by Ms. Surbhi Maheshwari; and Dr. Amit Kumar Sinhal on June 16, 2026, for A Spatio-Temporal Attention-Based System And Method For Isolated Sign Language Gesture Recognition.
Inventors include Ms. Surbhi Maheshwari; and Dr. Amit Kumar Sinhal.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: The present invention relates to a spatio-temporal attention-based system and method for isolated sign language gesture recognition. The system (100) comprises an input video module (110) configured to receive video clips in formats including .mp4 and .mov; a processing module (120) configured to decode the video into frames, uniformly sample T frames, resize, normalize pixel values, and generate a four-dimensional input tensor; a time-distributed spatial encoding module (130) that applies shared convolutional weights across frames to produce frame-level feature representations; a causal temporal modeling module (140) employing two causal one- dimensional convolutional blocks with 128 filters and kernel size 3 to preserve temporal order; a multi-head self-attention module (150) with four attention heads and key dimension 64 to refine temporal dependencies; a classifier module (160) comprising global average pooling and a dense layer with regularization; and an output module (170) generating gesture class probabilities. The system achieves high classification accuracy with low inference latency, enabling real-time assistive applications without requiring heavy computational infrastructure.
Disclaimer: Curated by HT Syndication.
