Questions about intrinsic calibration only

I reference the tutorial to calibrate a Sekonix SF3324 camera intrinsically.
But I have some questions below:

  1. I reference another topic and create a camera.front.json recorder configuration for record video step. This configuration contains the vehicle information as the attachment. Should I modify some vehicle information? or I can ignore this part, since this information won’t influence calibrate camera intrinsically.

  2. I also try the ros camera calibration before and use a 11x7 checkerboard like this ( I want to try the nvidia camera calibrate intrinsically too, but I found that the calibration process(calibration graph cli) need a targets.json, and all nvidia target boards are defined in it. Does this means I can’t do calibration by my original 11x7 checkerboard?

  3. Follow by question 2, if I could use the 11x7 checkerboard, where I should add the square size of
    checkerboard for calibration? How to add custom defined checkerboard information?

Dear cyan.chiu,

  1. AFAYK, the Intrinsics calibration is an internal calibration for the camera itself.
    So you can ignore the information about the car.

  2. Do you have any special reason to use the board you mentioned?
    If not, I’d like to recommend that you use the checkerboard we propose. Thanks.

Dear SteveNV,
The reason is the checkerboard images provided by dw1.2 data are not same to document description.
DW data:

The size just parameters to the calibration tool we can change as below command:

/usr/local/driveworks/tools/calibration/calibration-intrinsics-constraints --use-checkerboard=11x8 --input-video=/home/nvidia/camera-0.h264 --output=/home/nvidia/intrinsics/camera-0.json

And I trace the targets.json, I don’t find any checkerboard information in it. It contains the aprilTargets only, (correct me if I am wrong?) don’t influence calibration graph cli command tool to generate intrinsic validation images when I use other checkerboard.

So I think I don’t need to print another board for this calibration.
But I am curious about that ROS calibration tool need to set the square length parameters of checkerboard, I can’t find this parameters setting in NVIDIA calibration. If I miss something? or nvidia intrinsic calibration don’t need this information.

Dear cyan.chiu,

We give as an example a checkerboard of 10x7 pattern, and this is also what we show as test images in data/tools/calibration/intrinsic.
However, in the README file we write 11x8, which doesn’t fit the test data.
It is a mistake and README is not well aligned on the test data we show. Will fix it.

And you can use any checkerboard, just make sure to provide number of inner corners.
Better, to avoid any issues with specifying number of corners, you should use AprilTags instead. Thanks.

Dear SteveNV,
Thanks for your reply.
I am curious about that ROS calibration tool need to set the square length parameters of checkerboard.
I can’t find this parameters setting in NVIDIA calibration.
Does NVIDIA intrinsic calibration need this information?

Dear cyan.chiu,
Can I have this information somewhere in docs?

Dear SteveNV,
On the ros site(, I see the calibration need checkerboard square length as below image:

When I follow the NVIDIA camera calibration document, I doesn’t set this information.
So I want to know if I missed something or NVIDIA can detect it autonomously.

Dear cyan.chiu,

Thank you for sharing the info.
Please refer to below update for your topic. Thanks.
For intrinsic calibration size of the checkerboard doesn’t matter.
For extrinsic calibration it does matter, hence the correct size has to be added to target.json which lists the AprilTags target. Here is a cut-out from the documentation:

It is important that you modify this file and change the fields “barLength” for all targets which are present in the scene to the length in [meters] which you measured in Target Prints