You got me! I did oversimplify it.
There are 2 raycasts.
- The first goes between the main camera, through the hand, to a render texture (in Unity) that's on placed on a quad mesh (a 2d image).
- The second takes the position where the raycast hit the quad to then do a second raycast from another camera to where the actual background and target are being rendered.
Not quite analyzing an image, but it's another step. At first, I wanted to do a lot more with the green screen that may have needed this approach. Still, it removes the 3d perspective to make it more screen-like.