🧘 Posture Monitor

An offline, privacy-focused desktop application that uses AI to monitor your sitting posture in real-time and alerts you when you slouch.

✨ Features

• 🔒 100% Offline & Private: All processing happens locally on your device
• 🤖 AI-Powered Detection: Uses TensorFlow.js MoveNet for real-time pose tracking
• 📐 Personal Calibration: Adapts to your unique body type and sitting style
• ⚠️ Smart Alerts: Audio alerts after 3 seconds of bad posture
• 📊 Session Statistics: Track good vs. bad posture time
• 🎓 Personalized Training: Create a custom AI model trained on YOUR posture
• 🎨 Modern UI: Beautiful interface with live webcam feed and skeleton overlay

🚀 Quick Start

Prerequisites

• Node.js (v16 or higher) - Download here
• macOS (tested on macOS 10.15+) or Windows/Linux
• Webcam with permissions granted

Installation

1. Clone or download this repository

   git clone git@github.com:nttssv/posture_monitor.git
   cd posture_monitor

2. Install dependencies

   npm install

   Note: On first run, TensorFlow.js will download the AI model (~12MB). This requires internet connection only once. After that, the app works completely offline.

3. Add alert sound (included: beep.mp3)

   The repository includes a beep sound file. If you want to customize it, replace beep.mp3 with your own audio file (1-2 seconds recommended).

4. Run the application

   npm start

That's it! The app will launch automatically.

📖 How to Use

First Time Setup

1. Grant Camera Permission: When prompted, allow the app to access your webcam
2. Wait for AI Model: "Loading AI model..." will appear (10-30 seconds)
3. Position Yourself: Sit in front of the webcam in good lighting
4. Quick Calibration: Click "Calibrate Posture" while sitting with good posture

You're now monitoring your posture!

Understanding Posture

Good Posture Checklist:

• ✓ Head aligned over shoulders
• ✓ Shoulders relaxed and level
• ✓ Straight spine (natural curve)
• ✓ 90° angle at hips and knees
• ✓ Feet flat on floor

Common Bad Posture Mistakes:

• ✗ Forward head ("tech neck")
• ✗ Rounded shoulders
• ✗ Curved/hunched spine
• ✗ Leaning to one side

Using the Interface

Status Indicator (top-right overlay):

• ✅ Green = Good Posture
• ⚠️ Red = Bad Posture (with pulsing animation and audio alert)

Control Panel:

• Calibrate Posture: Quick baseline setup (5 seconds)
• Start Training: Create personalized AI model (2-3 minutes)
• Sound Alerts: Toggle audio on/off
• Log to CSV: Save posture data for analysis

Live Metrics:

• Shoulder angle deviation
• Neck angle deviation
• Overall posture score

Session Statistics:

• Time in good posture
• Time in bad posture
• Number of alerts triggered

🎓 Personalized Training (Recommended)

For best results, train a personalized model:

Why Use Training?

• Adapts to YOUR body: Everyone sits differently
• More accurate: Custom thresholds based on your posture
• Better alerts: Fewer false positives
• Learns continuously: Improves with feedback

How to Train

1. Click "Start Training" in the control panel

2. Capture Good Posture Samples (5-10 samples):

   • Sit in your ideal posture
   • Click "Capture Sample"
   • Move slightly between samples
   • Click "Next Step"

3. Capture Bad Posture Samples (5-10 samples):

   • Slouch or lean forward (intentionally bad)
   • Click "Capture Sample"
   • Vary your bad posture slightly
   • Click "Next Step"

4. Complete Training:

   • Click "Finish Training"
   • Your custom model is now active!

Training vs Quick Calibration

Feature	Quick Calibration	Personalized Training
Time	5 seconds	2-3 minutes
Samples	1	10-20
Accuracy	Good	Excellent
Learning	No	Yes (with feedback)
Best For	Quick setup	Daily use

Recommendation: Use Quick Calibration for immediate setup, then train a Personalized Model for best accuracy.

🔧 Configuration

You can adjust sensitivity in posture.js:

// Alert timing (lines 18-19)
this.alertDelay = 3000;      // Wait 3 seconds before alert
this.alertCooldown = 5000;   // 5 seconds between alerts

// Posture thresholds (lines 12-14)
this.shoulderAngleThreshold = 15;  // degrees
this.neckAngleThreshold = 10;      // degrees
this.deviationThreshold = 12;      // overall deviation

📁 Project Structure

poster_monitor_2/
├── main.js              # Electron main process
├── renderer.js          # Webcam & AI inference
├── posture.js           # Posture analysis & alerts
├── training.js          # Personalized training system
├── index.html           # UI layout
├── styles.css           # Modern UI styles
├── beep.mp3             # Alert sound
├── assets/              # Visual guides
│   ├── good-posture.svg
│   ├── bad-posture.svg
│   └── posture-comparison.svg
├── package.json         # Dependencies
└── README.md            # This file

🐛 Troubleshooting

Camera Not Working

• macOS: System Preferences → Security & Privacy → Camera → Enable for Electron
• Windows: Settings → Privacy → Camera → Allow apps to access camera
• Restart the app after granting permissions

Model Loading Failed

• Ensure internet connection on first run only
• Check console for errors: View → Toggle Developer Tools → Console
• Delete node_modules and run npm install again

Calibration Failed

• Ensure you're fully visible in the camera frame
• Improve lighting (face a window or turn on lights)
• Make sure head, shoulders, and hips are visible
• Try moving closer/farther from camera

No Pose Detected

• Check lighting conditions
• Ensure upper body is in frame
• Remove clutter from background
• Try different camera angles

Audio Not Playing

• Check system volume
• Verify beep.mp3 exists in project directory
• Test audio file separately
• Check DevTools console for errors

🔐 Privacy & Security

• No internet required after initial setup
• No data collection or analytics
• No external API calls
• All processing happens locally on your device
• Webcam feed never leaves your computer
• Training data stored locally in ~/Library/Application Support/PostureMonitor/

Your data never leaves your Mac!

💻 Development

Run with Developer Tools

npm run dev

Dependencies

• electron: Desktop app framework
• @tensorflow-models/pose-detection: AI pose estimation
• @tensorflow/tfjs-core: TensorFlow.js core
• @tensorflow/tfjs-backend-webgl: GPU acceleration
• @tensorflow/tfjs-converter: Model converter

Building for Distribution

npm install --save-dev electron-builder
npm run build

🎯 Tips for Best Results

1. Lighting: Ensure good, even lighting
2. Camera Position: Eye level, 2-3 feet away
3. Background: Clear, uncluttered background
4. Consistency: Use the same setup daily
5. Recalibrate: If you change desk/chair height
6. Take Breaks: Stand up and stretch every 30 minutes!

📊 Optional CSV Logging

Enable "Log to CSV" to save posture data:

• Logs every frame with timestamp
• Exports to posture_log_YYYY-MM-DD_HH-MM-SS.csv
• Useful for tracking progress over time

CSV Format:

timestamp,status,shoulderAngle,neckAngle,spineAngle,overallDeviation
2025-10-13T10:30:00.000Z,good,2.34,3.45,1.23,2.50

⚡ Performance

• CPU Usage: ~15-25% on modern MacBooks
• Memory: ~200-400MB
• GPU: Uses Metal (via WebGL) for acceleration
• Speed: 30-60 FPS real-time detection

📝 License

MIT License - Feel free to modify and distribute

🙏 Acknowledgments

• Built with Electron
• AI powered by TensorFlow.js
• Pose detection using MoveNet

🎉 Enjoy Better Posture!

Remember: This app is a tool to build awareness. Take regular breaks, stretch, and consult a healthcare professional for persistent posture issues.

---

Made with ❤️ for better ergonomics and health

Need Help?

1. Check the Troubleshooting section above
2. Open DevTools (View → Toggle Developer Tools) to see console logs
3. Ensure all dependencies are installed: npm install
4. Make sure your webcam is working in other apps first

Contributing

Found a bug or have a feature idea? Feel free to open an issue or submit a pull request!