Perspective n Point - RPnP algorithm

Question

I need to caculate the pose of a camera using an image of an artificial landmkark. For this porpouse I am trying to use the Perspective n Point approach so I can calculate it using the intrinsic camera matrix, the world coordinates of the landmark (I am using 4 points) and its projection in the image.

There are some algorithms to solve this (PnP, EPnP, RPnP, etc) and I am trying to use the RPnP. I have found an implementation of this here: http://xuchi.weebly.com/rpnp.html

I used this code but I am having some problems because I can't obtain the correct pose.

I am using the P.Corke's Robotics Toolbox for MATLAB to create a CentraCamera with a known pose and calculating the projection of the landmark in this camera, but the rotation and translation that the RPnP returns me is not the same as I defined before.

Anyone has used this RPnP algorithm to solve that kind of problems?

Answer 1

Since you only have 4 points, you should use the P3P algorithm. 3 points give you up to 4 solutions, and you need a fourth point to decide which one is correct. So 4 is the minimum number of points you need solve for the pose. Unfortunately, the P3P algorithm does not generalize to more points, which is why all those other algorithms you mentioned were invented.