I have a computer vision problem recostruction:
I have multiple view took with an uncalibrated camera (from wich i know to have fx = fy = f, s=0, and the parameters not varying)
The points that I can match from the different views all lay on a common plane. I can also match from the scene 4 different lines (lets say L1,L2,L3,L4)
such that: L1 is parallel with L2, L3 is parallel with L4 and L1 is orthogonal with L3 (and L4)...
I have enough views (9), and enough points (6 without the corner of the 4 lines).
I would like to use all the information i wrote to reduce noise, and I am asked to compute everything with the normalized images (that i don't know what does it means) to reduce numerical errors.
It is given also the distance between two of the matching points.
It is required to: reconstruct the scene (all the 6 3D points), reconstruct the camera internal parameter (ux, vx and f), and reconstruct the camera positions of the images.
So.
I studied the 8Point algorithm, but that algoritm (if i understood well) it is working with only two view, and more, it is not working with points laying on a plane.
I studied the 4Point algorithm, that is made for points laying on a same plane, but it requires only two views (and i would like to use all the views) and doesn't involves lines. Then other problem is that I don't understand how to reconstruct R and T (that are fundamental, if i understood well to reconstruct the position of the camera) from P.
Than i studied an iterative model that u can find at this paper:
3D Reconstruction with Uncalibrated Camera
but I'm not able (and i even don't know what is it) to perform rank 4 decomposition in matlab.
I don't ask you of course to give me a detailed solution of my problem, obvisiully, but of course I would like to understand at least wich way i should take to solve my problem in a correct way, then i can google more, and i can study more.
I thank you all for your time
