Exercise Predictor & Posture Corrector

A real-time exercise posture correction app that uses your webcam to detect exercises, count reps, and speak corrections aloud to help you maintain proper form.

Features

• Exercise Detection — Automatically identifies squats, push-ups, and planks from your webcam feed
• Rep Counting — Tracks repetitions with a state machine (and hold time for planks)
• Posture Analysis — 11 rule-based checks across 3 exercises using joint angle calculations
• Voice Feedback — Speaks corrections through your browser speakers with smart throttling (no repeated nagging)
• Visual Overlay — Color-coded skeleton (green/orange/red) with feedback text on the video feed
• EMA Smoothing — Reduces landmark jitter for stable angle readings

Supported Exercises & Posture Rules

Exercise	Rules Checked
Squat	Back straightness, knee valgus (caving in), squat depth, knees over toes
Push-up	Body alignment (sag/pike), elbow flare, range of motion, head position
Plank	Hip position (sag/pike), head alignment, elbow placement under shoulders

Tech Stack

• MediaPipe — Pose estimation (33 body landmarks)
• Streamlit — Web UI
• streamlit-webrtc — Real-time webcam streaming
• OpenCV — Frame processing and visual overlay
• Web Speech API — Browser-native text-to-speech for voice feedback

Setup

Prerequisites

• Python 3.10+
• Webcam
• A browser that supports Web Speech API (Chrome, Edge, Safari)

Installation

git clone https://github.com/alphoder/Exercise-Predictor-and-Posture-corrector.git
cd Exercise-Predictor-and-Posture-corrector
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Run

streamlit run app.py

The app opens in your browser. Click START, allow camera access, and begin exercising.

Usage Tips

• Position your camera from the side for best results with all three exercises
• Ensure your full body is visible in the frame
• Good lighting improves pose detection accuracy
• Use the sidebar to toggle voice feedback on/off or reset the rep counter

Project Structure

├── app.py                    # Streamlit UI, WebRTC streaming, main pipeline
├── pose_estimator.py         # MediaPipe PoseLandmarker wrapper + visual overlay
├── exercise_classifier.py    # Exercise detection + rep counting state machine
├── posture_analyzer.py       # Per-exercise posture correction rules
├── voice_feedback.py         # Throttled TTS via Web Speech API
├── angle_utils.py            # Joint angle geometry calculations
├── config.py                 # All thresholds and constants
└── requirements.txt          # Dependencies

How It Works

Each webcam frame flows through this pipeline (~20-35ms per frame):

Webcam Frame
  → MediaPipe Pose (33 landmarks)
    → Exercise Classifier (orientation + angle patterns → exercise type + rep state)
      → Posture Analyzer (joint angle rules → feedback list)
        → Visual Overlay (colored skeleton + text)
        → Voice Queue (throttled, priority-sorted browser TTS)

License

MIT