Skip to content
← Projects

ML / AI · CISS 2025, Published

Virtual Yoga Instructor with Real Time Feedback

A real-time pose estimation and corrective feedback system using normalized joint angle features and repetition counting. The system achieves sub-100 ms latency on standard hardware and remains robust to variations in body size and camera angle.

<100 ms

Latency

Norm. joint angles

Features

CISS 2025

Venue

Problem

Off-the-shelf pose estimation is sensitive to body size and camera angle, causing naive yoga correction systems to fail for most users.

Approach

I employed normalized joint angle features that are invariant to scale and orientation, combined with repetition-counting logic, and validated corrective feedback against reference pose deviations.

Results

The system achieved sub-100 ms latency on consumer hardware and remained robust across body sizes and camera angles without per-user retraining. The work was published at CISS 2025.

Stack

OpenCVMediaPipePyTorch