Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/elcorto/unfish
Correct fisheye distortions in images using OpenCV
https://github.com/elcorto/unfish
fisheye fisheye-lens-distortion fisheye-undistorting image-processing opencv
Last synced: about 1 month ago
JSON representation
Correct fisheye distortions in images using OpenCV
- Host: GitHub
- URL: https://github.com/elcorto/unfish
- Owner: elcorto
- License: bsd-3-clause
- Created: 2017-08-15T19:36:23.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2019-10-03T13:34:02.000Z (about 5 years ago)
- Last Synced: 2024-10-11T23:19:44.837Z (2 months ago)
- Topics: fisheye, fisheye-lens-distortion, fisheye-undistorting, image-processing, opencv
- Language: Python
- Size: 194 KB
- Stars: 30
- Watchers: 2
- Forks: 3
- Open Issues: 0
-
Metadata Files:
- Readme: README.rst
- License: LICENSE
Awesome Lists containing this project
README
unfish -- correct fisheye distortions in images using OpenCV
about
-----
This is basically a packaged and polished version of the OpenCV tutorial_ (see
also hack_) with a command line interface. It shows how to correct lens
distortions in images using OpenCV, based on chessboard calibration images
taken with the same camera.In my case, my mobile phone camera introduces a radial distortion (inverse
fisheye effect), hence the name.Here is an example of a distorted and corrected image.
.. image:: examples/fish.jpg
:width: 40%.. image:: examples/unfish.jpg
:width: 40%The script ``bin/unfish`` does all this and a little more::
usage:
unfish prep [-f ] (-p ...)
unfish calib [-r -f ] (-p ...)
unfish apply [-k ] ...commands:
prep optional preparation run, create rms_db.json
calib calibration run, calculate and write camera matrix and camera model
coeffs using chessboard calibration images to ./unfish_data
apply apply correction model to images, images are written to
./corrected_imagesoptions:
-p , --pattern-size size of the chessboard
(number of corners) in the calibration images, e.g. "9x6"
-f , --fraction fraction by which calibration files
have been scaled down (see bin/resize.sh)
-r , --max-rms in calibration, use only files with
rms reprojection error less than , uses rms_db.json
written by "prep"
-k keep that many path levels from , e.g.
files = /a/b/c/file1,/a/b/c/file2, and -k2, then store
./corrected_images/a/b/fileX instead of ./corrected_images/fileX [default: 0]In addition to the tutorial_, we added things like the ability to calculate the
RMS reprojection error per calibration image (``unfish prep``), in order to get
a feeling for the quality of the calibration per image.workflow
--------First, you print a chessboard and take a bunch of calibration images with the
affected camera, like this one:.. image:: examples/calib_pattern.jpg
:width: 40%Next, a calibration run will calculate correction parameters (camera matrix and
lens model coefficients, written to ``./unfish_data/``). Finally, you apply the
correction to all affected images. Corrected images are written to
``./corrected_images``.We found that it is a very good idea to scale down the chessboard calibration
images first. That makes the calibration part *a lot* faster (else the code
which searches for chessboard corners will run forever).Here is what you need to do, using a 9x6 chessboard.
::
$ ./bin/resize.sh 0.2 chess_pics/orig chess_pics/small
$ unfish calib -f 0.2 -p 9x6 chess_pics/small/*
$ unfish apply affected_pics/orig/*tips & tricks
-------------chessboard
You can grab a 7x7 chessboard image from the `OpenCV repo `_,
or a 9x6 from `older documentation `_. Remember: NxM are
the number of *corners*. It's hard to say how many calibration images you
need to take. We used around 100, but found that 5-10 good images have
basically the same effect. Also, make sure that the paper with the printed
chessboard is completely flat when you take photos.````
We found that excluding calibration images with a high per-image RMS
reprojection error (``unfish calib -r ...``) doesn't actually
improve the overall calibration, not sure why yet... _tutorial: http://docs.opencv.org/3.3.0/dc/dbb/tutorial_py_calibration.html
.. _hack: https://hackaday.io/project/12384-autofan-automated-control-of-air-flow/log/41862-correcting-for-lens-distortions
.. _chessboard: https://github.com/opencv/opencv/blob/master/samples/data/chessboard.png
.. _chessboard_old: http://docs.opencv.org/2.4/_downloads/pattern.pnginstall
-------
To let pip install all deps for you::$ git clone ...
$ pip3 install -e .