I have camera feed (in the form of RGB images) from 3 cameras with partially overlapping Field-of-view i.e. for the time stamp 0 to 100, I have total 300 frames or say synchronized 100 RGB frames for each camera.
Apart from this, the only available information is the Intrinsic Matrix which is same for all the cameras.
So, using these RGB frames from 3 cameras, how can I calibrate (extrinsic calibration) 3 cameras and calculate the global 3-DOF pose of an object (Robot - which is moving from one place to another)? For the object pose, assume a global coordinate system is defined by the first camera pose. (I might have access to the robot dimension: length, breadth, & height!)
If any direct method is not possible then I'm also open to using machine learning based method(s)?
Note: this question is an extension of my other question: How can I calibrate 3 cameras without knowing global pose of the object & camera locations? How can I find the pose of each camera wrt the first one?