TwoPointPosePartialRotation Steps

[alexs7]

Hello,

I am trying to get the solver work in a RANSAC loop.
According to http://theia-sfm.org/pose.html, the 3D model (or world) points have to be transformed before getting the image rays.

Before running the solver I have 2D-3D matches and the Rx and Rz rotations from the IMU.

1. Rotate the 3D world points by Rx and Rz (or Rz and Rx), to get 3D points in "image space".
2. Normalise the 3D points in "image space", to get the image rays.
3. Once the image rays are acquired, run the solver with input: axis = [0,1,0], 3D world points, and image rays.
4. The output pose is a word-to-camera (or image) space transformation.

The above was derived by also looking at the tests here.