FishEyeModel

Python project based on OpenCV module to model FishEye camera and undistort acquired images

Dependencies

Python 3.7 or higher
OpenCV 4.0 or higher

Calibration

OpenCV calibration algorithm for FishEye camera enables to model intrinsic and/or extrinsic parameters from a set of images acquired by the camera. To use the function, you have to acquire images of known pattern from which it is possible to extract distortions (chessboard for example). A chessboard pattern can be downloaded following this link: Chessboard pattern.

Then you have to acquire several images at different positions to integrate as much distorsion information as possible into the model. In our tests, a set of 25-30 images can be sufficient to have good undistort results.

We give a Python script to launch the calibration

python calibrate.py "ImageFolder" [-o "OutputFile"] [-s "True/False"]

ImageFolder (mandatory): folder where acquired images for calibration are stored.
OutputFile (optional): path of the saved model file. By default, file calibration.txt is saved into the current folder.
True/False (optional): boolean value to indicate if chessboard detection images have to be saved.

The main steps of the algorithm are:

1. Detection of all inner corners of the chessboard for each image.
2. Improvement of corner positions thanks to super-resolution algorithm.
3. Valid detections storage.
4. FishEye modelization on valid detections.
5. Save calibration parameters to file (extrinsic or intrinsic).

Note: calibration can failed even if chessboard detection is valid. Error can occured during the model computation (because of cv2.fisheye.CALIB_CHECK_COND flag). You have to remove the corresponding chessboard image to ensure good calibration.

Undistort

We give a Python script to launch the correction:

python undistort.py "SourceImage" [-c "CalibrationFile"] [-o "OutputImage"]

SourceImage (mandatory): path of the image to undistort.
CalibrationFile (optional): path of the model file generated by the calibration. If not set, the script will search the default file *calibratio.txt" into the current folder and fail if not found.
OutputImage (optional): path of the saved undistorted image. By default, undistorted image is saved in the file undistort.png into the current folder.