In order for the passive stereo to work, the images from the two separate projectors must match up on the screen. There are a few different approaches to this problem. Figure 3 illustrates four possibilities.
In our prototype system, we use projectors with integral shift lenses. These allow one to move the projected image several inches up or down, making it possible to match up the images from two stacked projectors. However, projectors with this feature are significantly more costly than those without it (by roughly 5000 per projector, at recent prices). The other arrangements in figure 3 are for cheaper projectors with fixed lenses.
The second approach is based on the fixed keystone offsets used in many projectors designed for meeting room environments, where they can be placed on a table, or mounted to the ceiling, and project onto a standard presentation screen. By mounting one projector on the ceiling, and the second at table height, the light cones will intersect at a certain distance, where the screen can be placed. The intersection point can be varied by changing the vertical distance between the two projectors.
The third approach is to simply throw away some image resolution. Two projectors can be stacked, and then only the area where their light cones overlap is used for the computer display. This means using the lower portion of the upper projector's image, and the upper portion of the lower projector's image; this involves using correspondingly offset window viewports in the graphics software configuration. With relatively slim projectors and a large screen, the lost image area can amount to less than 10% of the vertical resolution.
A final approach is to accept some slight distortion in the images. The two projectors can be stacked, but not exactly parallel to each other. By tilting one slightly, the two image areas on the screen will roughly overlap. The tilted projector's image will suffer some keystone distortion in this case, so the two images will not match exactly. But, some initial, non-rigorous tests suggest that the error is not significantly large, and may be acceptable to the average user.