1
$\begingroup$

thanks for taking the time to check my question. Firstly, let me describe the current configuration, I am using 4 cameras that produce non-overlapping images and the goal is to calculate their relative positions.

What I've tried so far:

[1] ROS Kalibr package. As the frames have a little overlap I tried to use a aprilgrid target hoping that if one marker can be caught to the frames I will have some hope to calibrate but it did not work, the overlapping areas are too small so there is no use to that.

[2] Meshroom camera rig calibration. For now this is my best bet but I can not find proper documentation about the workflow. Any ideas on how the pipeline should be are welcome.

In conclusion, all ideas are welcome because as you can see I have practically none.

P.S. I made this post more general instead of asking just about meshroom, thinking this might be helpful to more people in the future.

$\endgroup$
1
$\begingroup$

Not sure if there is an existing good sw for that but if you want to implement that a motion-based method will work.

The motion-based approach basically estimates the trajectory of each camera and finds extrinsic parameters(relative positions and rotations). It doesn't need any view overlap. What you need to do is simply grabbing your device and making a proper motion.

The simplest approach to implement the algorithm is using calibration boards. Attach multiple calibration boards in a room wall and estimate each camera trajectory by tracking the calibration board location using some library like OpenCV. Once you got the trajectory, the rest of the process is simple. See my paper if you want to know more theory.

If your cameras are installed on heavy stuff like a car it could be a bit tricky though.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.