Parallel rig
In a parallel rig, the optical axes of the two cameras are parallel to each other. Woods, Docherty and Koch (1993) recommend this arrangement because it produces no geometric distortion.
Parallel shooting without convergence
If the axes are parallel and the images are not shifted, parallax is zero only at infinity:
where: d parallax as a ratio of image width; the minus sign shows that the point is in front of the screen
The whole scene appears in front of the screen. This is why convergence is set separately in parallel shooting: either during the shoot with sensor shift, or in post-production with horizontal image translation (HIT).
Why parallel
- No keystone distortion: the two image planes lie in the same plane; no vertical parallax appears in the corners of the frame.
- The depth plane is flat: points at the same distance from the camera get the same parallax everywhere in the frame.
- Convergence can be chosen later: the zero parallax plane is decided during editing and mastering.
The cost of HIT: cropping
When convergence is set in post-production, the images are shifted in opposite directions. The total shift is:
where: C the desired convergence distance, N_px horizontal pixel count, Wc sensor width
At the edges of the shifted images, strips remain that belong to only one eye. The width of the common area becomes N_px − HIT; the image is either cropped to this width and scaled up, or a framing margin is left during the shoot.
Example: 65 mm interaxial, 3.5 m convergence. The total HIT is 50 pixels; 2.6% of the image width is cropped, and the remaining image is scaled up ×1.027.
In a virtual camera
In computer graphics, the parallel rig with shift is built with an asymmetric viewing frustum (off-axis projection). Because the virtual sensor is shifted without loss, no cropping is needed (NVIDIA 2011).