PNP settlement method (I may add more later when I have time)

Some experimental conclusions of pnp:
(1) Yaw angle stability:
In opencv,
SOLVEPNP_UPNP=SOLVEPNP_EPNP=SOLVEPNP_DLS>>SOLVEPNP_IPPE>SOLVEPNP_AP3P>SOLVEPNP_ITERATIVE
fixes the yaw angle of a recognition object check settlement. In this picture, l1 is ippe and l2 is AP3P , l3, l4 are UPNP and EPNP respectively, and they basically overlap.
The second picture l1, l2, l3, and l4 are IPPE, DLS, UPNP, and EPNP respectively. The latter three basically overlap and are regarded as equal, and there is no obvious data fluctuation.

1.EPnP

1.1 Control points

1. Formation: The homogeneous coordinates of a 3D reference point are linear combinations of the homogeneous coordinates of the control points
   

2. choose
   

3. Why are control points introduced: To improve the accuracy and stability of pose estimation. By introducing additional control points, the EPnP algorithm can better constrain the solution space in the process of estimating camera pose
   [Reference Zhihu] EPNP
   https://zhuanlan.zhihu.com/p/399140251