🚣‍♂️ Rowing Biomechanical Analysis System

AI-powered rowing technique analysis with real-time pose detection and force synchronization

🎯 Overview

This system provides comprehensive biomechanical analysis of rowing technique by combining:

• 🎥 Video capture with embedded timestamps
• 💪 PM5 force data capture and synchronization
• 🤖 AI pose estimation using YOLO11 with advanced smoothing
• 📊 Synchronized overlay videos with animated force curves and body angles
• 📈 Comprehensive reports with force-angle correlations

✨ Key Features

• Real-time pose detection with smooth, stable skeleton overlay
• Perfect synchronization between video, force data, and body angles
• Professional visualization with color-coded joint angle badges
• Comprehensive analysis including stroke-by-stroke breakdowns
• Export capabilities for further analysis in Excel, R, Python, or MATLAB
• Overlay-accurate force curves in reports (default mapping reproduces PM5 overlay)
• High-fidelity PM5 persistence (pm5_combined_strokes.json saved into analysis output)
• Per-stroke coaching metrics table embedded in comprehensive images

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/yourusername/rowingIA.git
cd rowingIA

# Install dependencies
pip install -r requirements.txt

# Install FFmpeg (required for video processing)
# macOS: brew install ffmpeg
# Ubuntu: sudo apt install ffmpeg

Usage

1. Connect your Concept2 PM5 via USB and start a workout
2. Capture data: sudo python py3row_usb_video_capture.py
3. Row for 2-5 minutes then stop capture
4. Analyze (auto-saves combined PM5 JSON; overlay mapping by default):
   • python create_complete_kinematics_overlay.py --session-dir py3rowcap_YYYYMMDD_HHMMSS
5. (Optional) Generate comprehensive images with the metrics table and overlay-mapped force curve:
   • python comprehensive_stroke_analysis.py analysis_py3rowcap_YYYYMMDD_HHMMSS
6. Review results in the generated overlay video, images, and report

📊 What You Get

Overlay Video Features

• 🟢 Smoothed skeleton overlay - Stable, professional pose detection
• 🎯 Joint angle badges - Real-time elbow, knee, hip, and ankle angles
• 📈 Animated force curves - Shows force throughout each stroke
• ⏱️ Perfect synchronization - Video and force data perfectly aligned

Analysis Reports

• Stroke-by-stroke breakdown - Duration, peak force, power, stroke rate
• Body angle statistics - Mean, standard deviation, min, max, range
• Force-angle correlations - How body position relates to force output
• Coaching metrics per stroke - Finish, Catch, and Sequence metrics shown in tables on images and summarized in report
• Detailed CSV data - For further analysis in your preferred tools

🛠️ Technical Details

AI & Computer Vision

• YOLO11 pose estimation - State-of-the-art pose detection
• Advanced smoothing algorithms - Eliminates jitter and bouncing
• Confidence-based filtering - Only uses high-quality detections
• Outlier rejection - Removes obviously incorrect poses

Data Processing

• Temporal smoothing - Averages keypoints over multiple frames
• Gaussian filtering - Reduces high-frequency noise
• Force curve analysis - Combines Drive + Dwelling phases
• Overlay mapping - Force curves in plots match the PM5 overlay exactly (drive-only, recovery zero)
• High-fidelity persistence - Saves pm5_combined_strokes.json and copies raw PM5 CSV into analysis folder

📁 Project Structure

rowingIA/
├── py3row_usb_video_capture.py          # Main capture script
├── create_complete_kinematics_overlay.py # Main analysis script
├── requirements.txt                      # Python dependencies
├── SETUP.md                             # Complete setup guide
├── py3rowcap_YYYYMMDD_HHMMSS/           # Capture session directory
└── analysis_py3rowcap_YYYYMMDD_HHMMSS/  # Analysis output directory

🎛️ Customization

Smoothing Parameters

Adjust pose smoothing for different stability/responsiveness:

# More stable (slower response)
PoseSmoother(window_size=7, confidence_threshold=0.6, outlier_threshold=2.5)

# More responsive (less stable)  
PoseSmoother(window_size=3, confidence_threshold=0.4, outlier_threshold=1.5)

Model Selection

Choose between different YOLO11 models:

• yolo11n-pose.pt - Nano (fastest, good accuracy)
• yolo11m-pose.pt - Medium (better accuracy, slower)
• yolo11l-pose.pt - Large (highest accuracy, slowest)

📋 Requirements

Hardware

• Concept2 PM5 rowing machine
• Computer with USB port
• Camera (webcam or external camera)
• USB cable to connect PM5 to computer

Software

• Python 3.8+
• FFmpeg for video processing
• See requirements.txt for Python dependencies

🔧 Troubleshooting

Common Issues

• No force data: Ensure PM5 is in workout mode, check USB connection
• Poor pose detection: Improve lighting, position camera for side view
• Permission denied: Run capture script with sudo for USB access

Performance Tips

• Faster processing: Use shorter video segments (2-5 minutes)
• Better accuracy: Ensure good lighting and camera positioning
• Stable results: Use consistent rowing pace

📚 Documentation

• Complete Setup Guide - Detailed installation and usage instructions
• Requirements - Python dependencies
• Analysis Examples - Sample output files

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

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

🙏 Acknowledgments

• Ultralytics for YOLO11 pose estimation models
• Concept2 for PM5 rowing machine communication
• OpenCV for computer vision capabilities
• Py3Row for PM5 data capture

---

Ready to analyze your rowing technique? 🚣‍♂️