I'am studiyng the problem of camera pose estimation in particular using ArUco marker and a monocular camera.
The documentations of ArUco and OpenCV are almost perfet, so the problem is not the algorith itself.
My problem is the camera and its choice.
I need to understand (maybe by a scientic studies analysis) wich camera params (eg. focal length, camera aperture, resolution, megapixels, etc.) are most influecing the precision and accuracy of the system, in particular since i have to reach a sub-millimetric accuracy of the detection.
There are many study in witch this result is reached (for example this study about a dodecapen), but they use an expensive camera and don't explain the camera features that are more important than others.
So, i can say important camera features for object detection (and pose estimation) are:
- Intrinsics params (focal length, principal point, skew wtc.) estimated with the calibration
- Lens distorsion factors
Morover my problem is a detection problem (i've only a photo) and not a tracking, even if there are many plus studies about tracking.
I would like to know for example if the camera of an Ipad is usable for my purposes: I know that there are many augmented reality works (using for example ARKit) on Ipad and Iphone, but what allow me to say that I can use an Ipad to reach a submilllimetric accuracy? ( i know that the response could be "tryng the camera in your system"... but i need to proof this thing by a theorical point of view..).
which params do I need to consider to choose the camera?
How can i say (without really use it) that a camera is o is not good?
Do you know articles, papers or other publications that can help me?
Thanks a lot!
