How to get points in stereo image from extrinsic parameters

Question

I've got a pair of cameras calibrated, knowing their intrinsic and extrinsic parameters. Knowing that they are both looking to a plane, if I define some points in one of the image, how can I get the point in the other image?

The cameras are pretty close one form the other, so suppose there is not any occlusion, both can see the same object.

Is there an openCV unction or set of functions to do this? My point is on Z=0 in the world.

Basically:

P_CAM1=(200,300) -> P_CAM2= ?

Crofton

code.google.com/p/sentience/wiki/StereoRanging I think this could be helpful. – Ahlgren

Answer 1

The answer:

1. Undistort images. cv::undistort
2. calculate P1ccdf = A^(-1)*P1 -> P1 point in f=1 world ref coords.
3. Calculate the point in the world ref Z=0 for that:
   • Calculate the optical centre coords in the world: Copt1=[0,0,0,1]T Copt1w=-R1t*T1
   • compute the vector between Coptw1 and P1ccdf: with some math -> V1w=R1T*P1ccdf
   • Coptw1+lambda*V1w=[Pxw,Pyw,0]T -> lambda=-cpotw(z)/V1(z)
   • Coptw1+lambda*V1w=[Pxw,Pyw,0]T=P1w
4. Protect world point into cam2
   • P1ccd2f = R2T2 *[Pxw Pyw 0 1]T
   • P1ccd2f=P1ccd2f/P1ccd2f(3)
   • P2=A2*P1ccd2f

where

• P means point

• A is the intrinsic matrix (4x4)

• RT is the camera matrix (3x4)

Crofton