📷 # 📷 Custom Pattern Camera Calibration-Tool with GA Optimization

A PyQt6-based desktop application for monocular camera calibration using custom (non-chessboard) patterns. The tool integrates a Genetic Algorithm (GA) optimizer that automatically selects the optimal subset of calibration images and OpenCV calibration flags, minimizing the reprojection error.

---

✨ Features

• 3 calibration modes in a unified GUI:
  • 📷 Real-Time: draw ROI live on the camera feed, capture frames and calibrate
  • 🖼️ Pattern Offline + Real-Time Photos: load a pattern image from disk, capture frames with the camera
  • 📁 Full Offline: load pattern + a set of images entirely from disk, no camera needed
• Automatic GA optimization after every calibration: no manual step required
• Custom pattern support via OpenCV ccalib (cv2.ccalib.CustomPattern)
• Results saved as .xml files (OpenCV FileStorage format)
• Per-mode GA log panel embedded in the UI, showing live epoch progress

---

🧬 How the GA works

After the standard cv2.calibrateCamera() call completes, the app automatically launches a GAWorker thread that runs a Genetic Algorithm on the full set of collected calibration frames.

Each chromosome encodes:

• A binary mask over the collected frames (which images to include)
• A set of OpenCV calibration flags (CALIB_FIX_K3, CALIB_RATIONAL_MODEL, etc.)

The fitness function runs cv2.calibrateCamera() on the active subset and computes the mean reprojection error. The GA evolves the population over multiple epochs using crossover, mutation, and a cataclysm mechanism to preserve diversity.

The best result is saved to out/ga_best_calibration.xml, containing:

• cameraMatrix, newCameraMatrix, distCoeffs
• ERR_REPROJ, N_IMAGES, ACTIVE_INDICES, FLAGS

---

📁 Project Structure

├── ccalib_ga.py            # Main entry point, MainWindow (PyQt6)
├── calibration_worker.py   # All 3 calibration mode widgets + workers (QThread)
├── ga_worker.py            # GAWorker QThread (wraps GAEngine)
├── ga_engine.py            # Genetic Algorithm core logic
├── ga_chromosome.py        # Chromosome representation (image subset + flags)
├── ga_constants.py         # GA hyperparameters
├── camera_calibrator.py    # CameraCalibrator: wraps cv2.calibrateCamera
├── pattern_manager.py      # PatternManager: wraps cv2.ccalib.CustomPattern
├── utils.py                # ChessboardValidator and utilities
└── res/
    └── camera_calib_tool_icon.png

---

🔧 Dependencies

Library	Purpose
opencv-python + opencv-contrib-python	Core computer vision, ccalib module
PyQt6	GUI framework
numpy	Numerical operations

Install standard dependencies:

pip install PyQt6 numpy

⚠️ OpenCV + ccalib on Windows 11 (Python 3.14)

The cv2.ccalib module is not included in the standard opencv-python pip package. To use it, OpenCV must be compiled from source with the opencv_contrib modules enabled.

Steps used to integrate ccalib on Windows 11 + Python 3.14.3:

1. Download OpenCV and opencv_contrib sources
2. Use CMake to configure the build, enabling only the required modules:
   • core, calib3d, ccalib, features2d, imgproc, highgui
3. Build with Visual Studio (or MSVC)
4. Place the compiled .pyd file — rename it to cv2.pyd — and all required .dll files into the Python site-packages directory:

C:\Users\<user>\AppData\Local\Programs\Python\Python314\Lib\site-packages\

5. Add or update __init__.py in that folder to expose the cv2 module correctly

After this, import cv2 and from cv2 import ccalib work normally in any Python script or virtual environment pointing to that interpreter.

---

🚀 Usage

python ccalib_ga.py

Real-Time mode

1. Click ▶ Start Camera
2. Draw the ROI rectangle around the calibration pattern in the video feed
3. Click ✅ Confirm ROI — the pattern is created
4. Click 📸 Capture Frame multiple times (≥ 10 recommended) from different angles
5. Click ⚙️ Perform Calibration + GA — standard calibration runs first, then the GA optimizer starts automatically in background

Offline mode

1. Click 📂 Load Pattern — select your pattern image
2. Click 🗂️ Load Calibration Images — select a batch of photos
3. Click ⚙️ Run Calibration + GA — the app processes all images, runs calibrateCamera, then launches the GA automatically

Results are saved to the out/ folder.

---

⚙️ GA Parameters

Tunable in ga_constants.py:

Parameter	Default	Description
GA_EPOCHS	25	Number of evolution generations
GA_POOL_SIZE	50	Population size
GA_P_MUTATION	0.25	Mutation probability
GA_P_CROSSOVER	0.75	Crossover probability
GA_T_CATACLYSM	0.01	Diversity threshold triggering cataclysm
GA_P_CATACLYSM	0.50	Fraction of population replaced on cataclysm
MIN_GENES	5	Minimum active images per chromosome
OPTIMIZE_FLAGS	True	Whether to also optimize calibration flags
PARALLEL	True	Run fitness evaluations in parallel threads

---

💡 Inspiration & References

• GA calibration image selection: ga_stereocalib by Alexandru-Ion Marinescu and A. Sergiu - IJCAI 2021 AI4AD Workshop by Alexandru-Ion Marinescu, Adrian-Sergiu Darabant and Tudor-Alexandru Ileni - genetic algorithm for stereo camera calibration image subset selection. This project adapts and extends that approach to monocular calibration with custom patterns.

• Custom Pattern Calibration: OpenCV ccalib tutorial (YouTube) and the official OpenCV ccalib module (cv2.ccalib.CustomPattern), part of opencv_contrib.

---

📄 License

This project is intended for research and educational use. (MIT)