The new camera stack of the systems Bullseye 32-bit and 64-bit does not work with OpenCV for video applications, for this you have to activate the old camera stack, but with mismatch of the function of the new library PiCamera2.
In principle, it is possible to configure the camera for the PiCamera2 library using a tuning_file (https://www.raspberrypi.com/documentation/computers/camera_software.html#more-about-libcamera). These files are .json files which allow to adjust the adjustable parameters for the specific camera model. The parameters have been determined specifically for each camera sensor, so that a manual calibration (chapter 6 https://datasheets.raspberrypi.com/camera/raspberry-pi-camera-guide.pdf) is normally not necessary.
If you do, you can consult the documentation of the parameters and experiment with them yourself.
The adjustment of the camera used with this tuning_file offers a lot of adjustment, but you cannot fix lens distortion like radial/tangential distortion with it. For this purpose, however, the special camera calibration using Camera Calibration can be used.
Thus it is theoretically possible to use the tuning parameters with PiCamera2 as well as the non-distortion of OpenCV for the camera recording.
| PiCamera2 | OpenCV | PiCamera2&OpenCV |
|---|---|---|
| - use the .json tuning_file with a lot of algorithms (12) for RPi cameras | - Enable the old camera-stack for RPi and use the camera calibration for undistortion with the OpenCV provided algorithms (16) (more than PiCamera2) | - works, but the stream is very slow (suggestions for improvement welcome github_calibtest.py ir_cut_picamera2_array.py) |
If you intend to calibrate your camera independently of PiCamera2 due to strong distortion, I provide the following scripts. The first script is used for capturing with a 5 second timer, which is basically used for creating the images for the camera calibration.
Take and save pictures with self-triggered timer:
ir_cut_picamera2_timer.py: This second script loads the calibration images of the default folder "calib_images" you have taken with the ir_cut_picamera2_timer.pyscript. Further you must give the folder, where the undistorted images after calibration get saved. The last argument is the board dimension, which must be given correctly, because many people make a mistake here, which causes that the algorithms can't find all corners and return False for some calibration images.
python3 calibrate_camera.py --imgdir=calib_imabes --savedir=undistorted_images --board=9x6The script needs the destination, where the calibration images for OpenCV Camera Calibration get saved. Additionally you can adjust the time before a picture is taken, to position the chessboard before taking the image. To achieve a sufficient accuracy it's recommended to take between 10-20 (or more) images of the chessboard.
sudo python3 ir_cut_picamera2_timer.py --imgdir=calibration_images --res=1920x1080 --time=5Note: Further information here Camera Calibration.