Workout form analysis model preview

Workout Form Analysis Model

From idea → to real-time system → to iteration

I built a computer vision system that analyzes workout technique using angle calculations. It tracks body landmarks, detects incorrect posture, counts reps, and provides real-time feedback. When form becomes unsafe, the system triggers an audio warning.

Why it matters

Many people train without a coach. Small mistakes in posture can lead to long-term injuries. This system focuses on catching those mistakes early and helping users fix them instantly.

Validation

The system and approach were validated through feedback from two professionals in the computer vision field (with 6+ and 10+ years of experience), leading to improvements in evaluation logic and robustness.

Watch it in action

See how the system detects mistakes in real time.

How I built it Main Article

A full breakdown of the system, including logic, architecture, and design decisions.

Read full article →

🎥 Video breakdown

A walkthrough of how the system works, including code and logic.

Watch video →

📄 First version

My initial attempt — useful to see how the system evolved.

View first version →

Tech stack

  • Python
  • MediaPipe
  • OpenCV
  • NumPy

How it works

  • Detects body landmarks using MediaPipe Pose
  • Calculates joint angles (knees, hips, ankles)
  • Compares them to safe thresholds
  • Detects incorrect posture in real time
  • Triggers feedback and sound alerts
  • Counts reps and guides full workout
  • Provides final summary and feedback

Project Demo — Version 1