Skip to content

Open source software for automatic image calibration from a set of images that are manually calibrated.

License

Notifications You must be signed in to change notification settings

jesogon/UCalib

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

91 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

UCalib

UCalib is an open source software written in Python for automatic image calibration from a set of images that are manually calibrated.

Description

The calibration algorithm assumes that the camera position and the intrinsic parameters of the camera remain unchanged. The result of the process is a common position and intrinsic camera parameters for all images, and the orientation of the cameras for each of the images. In addition, planviews can be generated for each image. The development of this software is suitable for Argus-type video monitoring stations. Details about the algorithm and methodology are described in

Simarro, G.; Calvete, D.; Souto, P. UCalib: Cameras Autocalibration on Coastal Video Monitoring Systems. Remote Sens. 2021, 13, 2795. https://doi.org/10.3390/rs13142795

The automatic calibration process consists of two steps:

  1. Basis calibration
  2. Automatic image calibration

Further UCalib allows to generate planviews for the calibrated images:

  1. Planview generation

A code to verify the quality of the GCPs used in the manual calibration of the basis images is also provided:

  1. Check GCP for basis calibration

Requirements and project structure

To run the software it is necessary to have Python (3.8) and install the following dependencies:

  • cv2 (4.2.0)
  • numpy (1.19.5)
  • scipy (1.6.0)

In parenthesis we indicate the version with which the software has been tested. It is possible that it works with older versions.

The structure of the project is the following:

  • example.py
  • example_notebook.py
  • ucalib
    • ucalib.py
    • ulises_ucalib.py
  • example
    • basis
      • basisImage01.png
      • basisImage01cal0.txt
      • basisImage01cal.txt
      • basisImage01cdg.txt
      • basisImage01cdh.txt
      • . . .
    • basis_check
      • basisImage01.png
      • basisImage01cdg.txt
      • . . .
    • images
      • image000001.png
      • image000001cal.txt
      • . . .
    • planviews
      • crxyz_planview.txt
      • xy_planview.txt
      • image000001plw.png
      • . . .
    • TMP
      • basisImage01cal0_check.png
      • basisImage01cal_check.png
      • image000001cal_check.png
      • image000001_checkplw.png
      • image000001plw_check.png
      • . . .

The local modules of UCalib are located in the ucalib folder.

To run the demo in the folder example with the basis of images in basis and the images in images using a Jupyter Notebook we provide the file example_notebook.ipynb. For experienced users, the example.py file can be run in a terminal. UCalib handles PNG (recommended) and JPEG image formats.

Basis calibration

To manually calibrate the images selected for the basis, placed in the folder basis, it is necessary that each image <basisImage>.png is supplied with a file containing the Ground Control Points (GCP) and, optionally, a file with the Horizon Points (HP). The structure of each of these files is the following:

  • <basisImage>cdg.txt: For each GCP one line with (minimum 6)

pixel-column, pixel-row, x-coordinate, y-coordinate, z-coordinate

  • <basisImage>cdh.txt: For each HP one line with (minimum 3)

pixel-column, pixel-row

Quantities must be separated by at least one blank space between them and the last record should not be continued with a newline (return).

To generate <basisImage>cdg.txt and <basisImage>cdh.txt files the UClick software is available.

Run basis calibration

Import modules:

import sys
import os
sys.path.insert(0, 'ucalib')
import ucalib as ucalib

Set the main path and the path where the basis is located:

pathFolderMain = 'example'
pathFolderBasis = os.path.join(pathFolderMain, 'basis')

Set the value of maximum error allowed for the basis calibration:

Parameter Suggested value Units
Critical reprojection pixel error eCritical 5. pixel
eCritical = 5.

Select an intrinsic camera calibration model.

Camara model parabolic quartic full
Lens radial distortion parabolic parabolic + quartic parabolic + quartic
Lens tangential distortion no no yes
Square pixels yes yes no
Decentering no no yes

The parabolic model is recommended by default, unless the images are highly distorted.

calibrationModel = 'parabolic'

In case certain variables of the calibration model (see table below) are to be fixed, the values have to be provided in the givenVariablesDict dictionary. For a full calibration, the dictionary has to be left empty, i.e. givenVariablesDict = {}.

givenVariablesDict = {'zc':142.5, 'k1a':-0.0025}

To facilitate the verification that the GCPs have been correctly selected in each image of the basis, images showing the GCPs and HPs (black), the reprojection of GCPs (yellow) and the horizon line (yellow) on the images can be generated. Set parameter verbosePlot = True, and to False otherwise. Images (<basisImage>cal0_check.png) will be placed on a TMP folder.

verbosePlot = True

Run the initial calibration algorithm for each image of the basis:

ucalib.CalibrationOfBasisImages(pathFolderBasis, eCritical, calibrationModel, givenVariablesDict, verbosePlot)

In case that the reprojection error of a GCP is higher than the error eCritical for a certain image <basisImage>, a message will appear suggesting to re-run the calibration of the basis or to modify the values or to delete points in the file <basisImage>cdg.txt. If the calibration error of an image exceeds the error eCritical the calibration is given as failed. Consider re-run the calibration of the basis or verify the GPCs and HPs.

Then, run the algorithm to obtain the position and the optimal intrinsic parameters of the camera:

ucalib.CalibrationOfBasisImagesConstantXYZAndIntrinsic(pathFolderBasis, calibrationModel, givenVariablesDict, verbosePlot)

As a result of the calibration, the calibration file <basisImage>cal.txt is generated in the basis directory for each of the images. This file contains the following parameters:

Magnitudes Variables Units
Camera position coordinates xc, yc, zc m
Camera orientation angles ph, sg, ta rad
Lens radial distortion (parabolic, quartic) k1a, k2a -
Lens tangential distortion (parabolic, quartic) p1a, p2a -
Pixel size sc, sr -
Decentering oc, or pixel
Image size nc, nr pixel
Calibration error errorT pixel

The different calibration files <basisImage>cal.txt differ only in the angles of the camera orientation (ph, sg, ta) and the calibration error (errorT). A <basisImage>cal0.txt file with the initial calibration parameters for each image of the basis will also have been generated.

Automatic image calibration

In this second step, each of the images in the folder images will be automatically calibrated. Set the folder path where images to calibrate automatically are stored. To facilitate the verification of the calibration of each image, images showing the reprojection of the GCPs and the horizon line can be generated. Set parameter verbosePlot = True, and to False otherwise. Images(<images>cal_check.png) will be placed on a TMP folder. In the case that the images in the folder images have already been calibrated, set overwrite = True to generate them again and to False otherwise.

pathFolderImages = os.path.join(pathFolderMain, 'images')
overwrite = False
verbosePlot = True

Set the values of the automatic image calibration parameters:

Parameter Suggested value Units
Number of features to identify with ORB nORB 10000 -
Critical homography error fC 5. pixel
Critical number of pairs KC 4 -
nORB, fC, KC = 10000, 5., 4

Run the algorithm to calibrate images automatically:

ucalib.AutoCalibrationOfImages(pathFolderBasis, pathFolderImages, nORB, fC, KC, overwrite, verbosePlot)

For each of the images <image>.png in directory images, a calibration file <image>cal.txt with the same characteristics as the one described above will be obtained. The autocalibration process may fail because the homography error is higher than the one set by the parameter fC, the number of pairs is lower than the critical value KC or ORB not being able to identify pairs in the image. In any of these cases it is reported that the calibration of the image <image>.png has failed.

Planviews

Once the frames have been calibrated, planviews can be generated. The region of the planview is the one delimited by the minimum area rectangle containing the points of the plane specified in the file xy_planview.txt in the folder planviews. The planview image will be oriented so that the nearest corner to the point of the first of the file xy_planview.txt will be placed in the upper left corner of the image. The structure of this file is the following:

  • xy_planview.txt: For each points one line with

x-coordinate, y-coordinate

A minimum number of three not aligned points is required. These points are to be given in the same coordinate system as the GCPs.

Set the folder path where the file xy_planview.txt is located and the value of z0.

pathFolderPlanviews = os.path.join(pathFolderMain, 'planviews')
z0 = 3.2

The resolution of the planviews is fixed by the pixels-per-meter established in the parameter ppm. To help verifying that the points for setting the planview are correctly placed, it is possible to show such points on the frames and on the planviews. Set the parameter verbosePlot = True, and to False otherwise. The images (<image>_checkplw.png and <image>plw_check.png) will be placed in a TMP folder. In the case that planviews have already been generated, set overwrite = True to generate them again and to False otherwise.

ppm = 2.0
overwrite = False
verbosePlot = True

Run the algorithm to generate the planviews:

ucalib.PlanviewsFromImages(pathFolderImages, pathFolderPlanviews, z0, ppm, overwrite, verbosePlot)

As a result, for each of the calibrated images <image>.png in folder images, a planview <image>plw.png will be placed in the folder planviews. Note that objects outside the plane at height z0 will show apparent displacements due to real camera movement. In the same folder, the file crxyz_planview.txt will be located, containing the coordinates of the corner of the planviews images:

  • crxyz_planview.txt: For each corner one line with

pixel-column, pixel-row, x-coordinate, y-coordinate, z-coordinate

GCP check

To verify the quality of the GCPs used in the manual calibration of the basis images, a RANSAC (RANdom SAmple Consensus) is performed. Points of the files <basisImage>cdg.txt located at the basis_check folder will be tested. The calibration of the points (minimum 6) is done assuming a parabolic camera model and requires the maximum reprojection pixel error eCritical for the GCPs. Set the folder and run the RANSAC algorithm:

pathFolderBasisCheck = os.path.join(pathFolderMain, 'basis_check')
ucalib.CheckGCPs(pathFolderBasisCheck, eCritical)

For each file <basisImage>cdg.txt, the GCPs that should be revised or excluded will be reported.

Contact us

Are you experiencing problems? Do you want to give us a comment? Do you need to get in touch with us? Please contact us!

To do so, we ask you to use the Issues section instead of emailing us.

Contributions

Contributions to this project are welcome. To do a clean pull request, please follow these guidelines.

License

UCalib is released under a GPLv3 license. If you use UCalib in an academic work, please cite:

@Article{rs13142795,
  AUTHOR = {Simarro, Gonzalo and Calvete, Daniel and Souto, Paola},
  TITLE = {UCalib: Cameras Autocalibration on Coastal Video Monitoring Systems},
  JOURNAL = {Remote Sensing},
  VOLUME = {13},
  YEAR = {2021},
  NUMBER = {14},
  ARTICLE-NUMBER = {2795},
  URL = {https://www.mdpi.com/2072-4292/13/14/2795},
  ISSN = {2072-4292},
  DOI = {10.3390/rs13142795}
  }

@Online{ulisesdrone, 
  author = {Simarro, Gonzalo and Calvete, Daniel},
  title = {UCalib},
  year = 2021,
  url = {https://github.com/Ulises-ICM-UPC/UCalib}
  }

About

Open source software for automatic image calibration from a set of images that are manually calibrated.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 88.9%
  • Jupyter Notebook 11.1%