MUMBAI, India, June 16 -- Intellectual Property India has published a patent application (202611053681 A) filed by Ms. Deepali Dev; Mr. Abdullah Masood; and Mr. Altaf Hussain, Ghaziabad, Uttar Pradesh, on April 27, for 'yoga one: an ai-powered real-time yoga posture correction system.'

Inventor(s) include Ms. Deepali Dev; Mr. Abdullah Masood; and Mr. Altaf Hussain.

The application for the patent was published on June 5, under issue no. 23/2026.

According to the abstract released by the Intellectual Property India: "Yoga practice without an instructor often goes wrong in ways the practitioner cannot see. The work described here addresses that gap through Yoga One, a vision-driven assistant that watches the user, decides which asana is being attempted, and points out what needs to change. The pipeline begins with a 2D landmark estimator that locates the major joints from a single webcam frame. Those raw points are not used directly. Instead, they are reshaped into a compact descriptor built from three ingredients: coordinates that have been re-anchored on the hip midpoint and divided by torso length; angles at the elbows, knees, and hips computed through dot products; and ratios of distances between selected joint pairs. Each value is scaled by the confidence the estimator reports for the underlying landmark, so frames with weak detections contribute less. A small one-dimensional convolutional model takes this descriptor and outputs a pose label along with a confidence number. From there, the angle of every relevant joint iscompared against a stored reference for that pose, and the gaps drive a short list of textual prompts suchas "straighten the back" or "ease the knee bend." On a curated set of ten common asanas, the prototypereaches 93.7 percent classification accuracy and runs at 15 to 20 frames per second on an Intel i5laptopwith 8 GB of RAM, which is comfortable enough for live coaching. Because nothing in the stack dependson depth cameras, IMUs, or wearables, the whole arrangement can be rolled out on whatever cameratheuser already owns. Keywords: Yoga, Pose Estimation, Deep Learning, Posture Correction, Computer Vision, MediaPipe, Joint Angle Analysis, 1D Convolutional Neural Network, Confidence-Weighted Features, Real-TimeFeedback, Scale-Invariant Features, Human Pose Estimation.''

Disclaimer: Curated by HT Syndication.