Calimiro AI - Real-time Exercise Detection

A computer vision system for real-time exercise classification and repetition counting using MediaPipe pose detection and deep learning models.

🏗️ Project Structure

SmartMirror/
├── 📁 src/                        Source code
│   ├── 📁 backend_interface/
│   │   ├── backend_interface.py   Backend interface to communicate with the mirror
│   │   ├── server.py              HTTP server to receive data batches from the mirror
│   │   └── shared_data.py         Thread-Safe IPC
│   ├── 📁 core/                   Core functionality
│   │   ├── realtime_pipeline.py   Main realtime pipeline
│   │   ├── dataset_builder.py     Dataset creation and management
│   │   ├── realtime.py            Realtime model & OpenCV rendering
│   │   ├── realtime_no_counting.py
│   │   ├── main.py                Main script
│   │   └── __init__.py
│   ├── 📁 training/               Model training
│   │   ├── train_classifier.py    Classification model training
│   │   ├── retrain_segmentation_realtime.py   Real-time segmentation training
│   │   ├── classification/        Classification modules
│   │   └── segmentation/          Segmentation modules
│   ├── 📁 testing/                Testing and debugging
│   │   ├── test_*.py              Various test scripts
│   │   ├── debug_classification.py
│   │   └── inspect_classification_dataset.py
│   ├── 📁 utils/                  Utility scripts
│   │   ├── classification_demo.py
│   │   ├── segmentation_demo.py
│   │   ├── mediapipe_demo.py
│   │   └── video_labeler.py
│   └── __init__.py
├── 📁 data/                       Data storage
│   ├── 📁 raw/                    Raw video files
│   ├── 📁 processed/              Processed datasets
│   └── 📁 labels/                 Annotation files
├── 📁 models/                     Trained models
│   ├── 📁 classification/         Classification models
│   └── 📁 segmentation/           Segmentation models
├── 📁 logs/                       Training logs
├── 📁 output_videos/              Output videos
├── 📁 test_videos/                Test videos
├── 📁 docs/                       Documentation
├── 📁 scripts/                    Utility scripts
├── 🐍 run_realtime.py             Main entry point
├── 📄 requirements.txt            Dependencies
└── 📄 README.md                   This file

🚀 Quick Start

Installation

1. Clone the repository:

   git clone <repository-url>
   cd SmartMirror

2. Install dependencies:

   pip install -r requirements.txt

3. Run the real-time pipeline:

   python run_realtime.py

Usage

• Press 'q' to quit the application
• Press 'r' to reset rep counts
• The system will automatically detect and classify exercises
• Rep counts are displayed in real-time

🧠 Features

✅ Exercise Classification

• 4 Exercise Types: Push-ups, Squats, Pull-ups, Dips
• Real-time Detection: Live webcam feed processing
• High Accuracy: 99.98% test accuracy
• Low Latency: Optimized for real-time performance

✅ Rep Counting

• State Machine: Advanced rep detection algorithm
• Low-FPS Optimized: Works with 3 FPS performance
• Multiple Exercises: Simultaneous tracking
• Visual Feedback: Real-time state display

✅ Performance Optimizations

• 30-Frame Windows: Optimized for real-time processing
• Frame Skipping: Reduces computational load
• Adaptive Thresholds: Adjusts to performance constraints
• Efficient Models: Lightweight neural networks

🔧 Development

Training Models

Classification Model:

cd src/training
python train_classifier.py

Real-time Segmentation Models:

cd src/training
python retrain_segmentation_realtime.py

Testing

Test with video files:

cd src/testing
python test_realtime_video.py test_videos/test0.mp4

Debug classification:

cd src/testing
python debug_classification.py test_videos/test0.mp4

📊 Performance

• Classification Accuracy: 99.98%
• Real-time FPS: ~3 FPS (CPU-based)
• Latency: ~350ms per frame
• Memory Usage: ~2GB RAM

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• MediaPipe: Pose detection framework
• TensorFlow: Deep learning framework
• OpenCV: Computer vision library