MultiCalib 0.1.1

A Python library for multi-camera system calibration with high accuracy.

MultiCalib

MultiCalib is a Python library designed to perform multi-camera system calibration on 4k high-resolution images. This library allows for individual camera calibration and full combined calibration optimization using different patterns. MultiCalib supports high-resolution 4k images with reprojection errors as low as 0.8 pixels. The more patterns used in the calibration process, the lower the reprojection error, ensuring high accuracy and precision in the calibration results.

Installation

You can install MultiCalib via pip:

pip install MultiCalib

Requirements

• matplotlib==3.7.5
• networkx==3.1
• numpy==1.23.5
• opencv-python==4.7.0.72
• scikit-learn==1.3.0
• scipy==1.10.1
• PyYAML==6.0

Classes

CameraSystemCalibration

This class holds information about all the cameras to be calibrated and provides methods to perform full combined calibration optimization.

Methods

• __init__(self, data, patternSize, squareSize, minNoPatterns=10, optimizeSingleCalib=True, drawGraph=False): Initializes the camera system calibration.
• calibrateSystem(self): Performs bundle adjustment on all cameras together for multicamera calibration.
• exportCameraParameters(self, calibrationPath): Saves the camera parameters to the specified path. reprojectPoints(self, outputPath): Saves the images with the reprojected points.

Camera

This class holds parameters for each individual camera.

Pattern

This class holds information about all the patterns used in the calibration process.

Usage

Here's an example use case of the MultiCalib library:

import os
from glob import glob
from MultiCalib import CameraSystemCalibration

inputPath = './calib/'
outputPath = './'
patternSize = [(12, 11), (14, 13)]
squareSize = [0.035, 0.042]

data = {}
cameraList = sorted(os.listdir(inputPath))
calibrationPath = f'{outputPath}/camera_parameters.yaml'

for camera in cameraList:
    cameraPath = os.path.join(inputPath, camera)
    calibrationImagesPath = sorted(glob(os.path.join(cameraPath, '*.png')))
    data[camera] = calibrationImagesPath

CameraSystem = CameraSystemCalibration(data, patternSize, squareSize, minNoPatterns=10, optimizeSingleCalib=True)
CameraSystem.calibrateSystem()
CameraSystem.exportCameraParameters(calibrationPath)
CameraSystem.reprojectPoints(outputPath)

Data Structure

The data dictionary should be created with the key being the camera, and the value being a sorted list of the image paths.

data = {
    'camera1': ['path/to/image1.png', 'path/to/image2.png', ...],
    'camera2': ['path/to/image1.png', 'path/to/image2.png', ...],
    ...
}

Detailed Explanation

1. Initialization:
   • CameraSystem = CameraSystemCalibration(data, patternSize, squareSize, minNoPatterns=10, optimizeSingleCalib=True) initializes the system and performs individual camera calibration without extrinsics.
   • patternSize is a list of tuples that contains the dimensions of the patterns used.
   • squareSize is a list of the length of one side of the squares in each pattern (measured in meters).
   • optimizeSingleCalib is True if you want to run bundle adjustment optimization on the single camera calibration (default is False).
2. System Calibration:
   • CameraSystem.calibrateSystem() performs bundle adjustment on all cameras together for multicamera calibration.
3. Export Camera Parameters:
   • CameraSystem.exportCameraParameters(calibrationPath) saves the camera parameters in the specified path.
4. Reproject Points:
   • CameraSystem.reprojectPoints(outputPath) saves the images with the reprojected points.

Accessing Camera and Pattern Information

• To view a camera's pose or intrinsic parameters:

  CameraSystem.cameras[0].intrinsic
  CameraSystem.cameras[0].pose
  CameraSystem.cameras[0].distortion
  

• To view pattern information for a camera:

  CameraSystem.cameras[0].patterns[0].pose
  CameraSystem.cameras[0].patterns[0].imgPoints
  CameraSystem.cameras[0].patterns[0].objPoints
  

Contributing

Contributions are welcome! Please submit a pull request or open an issue to discuss changes.

License

This project is licensed under the MIT License. Copyright (c) 2024 Obumneme Dukor

Citation

Please cite our work if you use MultiCalib.

@article{Dukor2024,
    author    = {Obumneme Dukor and S. Mahdi H. Miangoleh and Yağız Aksoy},
    title     = {{Multicalib}: A Modern Library for High Resolution Multicamera Calibration},
    journal   = {arXiv:XXXX.XXXXX},
    year      = {2024},
}

References

• Byrd, R. H., Schnabel, R. B., & Shultz, G. A. (1988). Approximate solution of the trust region problem by minimization over two-dimensional subspaces. Mathematical Programming, 40, 247–263. https://doi.org/10.1007/BF01580735
• Li, B., Heng, L., Köser, K., & Pollefeys, M. (2013). A Multiple-Camera System Calibration Toolbox Using A Feature Descriptor-Based Calibration Pattern. 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November 3-7, 2013. Tokyo, Japan. https://doi.org/10.1109/IROS.2013.6696517