Inaccurate estimation of YAW, PITCH and ROLL of camera poses

My problem is a bit more general in nature.
I need to estimate Roll, Pitch and Yaw of a camera, specifically we are talking about arducam.
The sensor of the camera unfortunately by construction is slightly inclined, so I need to try to estimate the angles of inclination.

First I used Aruco marker, the estimation is good, but it is necessary that in the scene there are at least 3 Aruco to have a good estimate of the position.
My goal though remains to estimate the first position and then use the regularity of the structure to get a correct camera position estimation.

With that we realized that the problem remains in estimating the camera tilt.

So we fixed a chessboard in the ground where the structure lies with arducam placed in the arm, but unfortunately the estimation is not accurate and often is not detected due to its position.
The configuration is shown in the image below, we also have an encoder that defines a precise position for the arm angle.

Since this is my first time trying to estimate extrinsics camera pose, do you have any tips or improvements to try to get as accurate an estimate as possible?


Estimating pose can be done with a simple function, or can need a huge work and knowledge. It depends on the particular case.

So, to know more about your particular case, It would be better to share some real photos of your 1 DOF robot, and some camera captures from your many scenarios: ARUCO, chessboard, none, etc.