Some people have no stereo vision and need monocular depth-cue fallback

What it is

People who had strabismus, amblyopia, or unused binocular vision in the developmental window often grow up stereoblind or stereoanomalous: a shape hidden in a random-dot stereogram is noise to them; the disparity channel never came online. This is not a rounding error in the population — screening studies commonly land between several percent and about one in ten. They walk and grasp on occlusion, perspective, shadow and motion parallax. An interface that puts front-to-back solely in the interocular difference hands them a flat picture.

This is not “disparity only works near”. They do not have the channel even near.

Why it happens

Stereo matching is learned in a developmental window. If the two eyes’ images do not line up in that window, cortex does not build disparity detectors. Adulthood plus a headset still produces a left and a right image; that layer of computation has no audience. They are not necessarily “unable to see sharply” — acuity can be normal — the two images simply contribute nothing to depth.

Monocular cues remain. Occlusion is almost non-negotiable: near covers far. Contact shadows declare which surface an object sits on. Texture gradients, relative size, and the motion parallax of a small head movement can hold the reachable range together. If a headset drops those and keeps only a stereo “pop”, stereoblind wearers read every layer as pasted on one board, and clicks and grabs land with 2D habits — clipping and empty grasps are that reading’s consequences.

The design inference is not “ship a stereoblind mode”. It is: the scene’s depth should still be readable monocularly by default.

Studying it

Screen with Titmus, Randot, or random-dot tests into stereo-normal versus stereo-anomalous, then run the same depth tasks: ordering, grasping, clicking front versus back panels. One condition keeps all monocular cues; another strips them with random dots or by removing shadows and occlusion.

Independent variables: stereo screening result, whether monocular cues are retained, whether the task demands absolute distance. Dependent variables: depth-order accuracy, miss rate on grasps, completion time, reports of “everything is equally far”.

Recruiting only from lab colleagues systematically misses stereo anomaly, and the product will treat stereo as a universal channel.

Where it stops holding

Temporarily covering one eye (injury, a fogged lens, a dead display) produces the same monocular state, but it is acute and people know they are “one eye down”; congenital stereoblindness need not come with that metacognition, and nobody goes looking for a “stereo switch”. Some stereoanomaly still has coarse disparity without fine; widely separated layers may sort, stacked fine offsets will not. In video see-through the real world already supplies rich monocular cues; virtual objects that fight real ones on occlusion fail more loudly than in a fully virtual scene. Screening plates are high-contrast contours; semi-transparent panels on a product are harder to fuse, so anomaly will show up more often in the wild than in the lab.

Applying it

• Build front-to-back with occlusion, contact shadows and relative size first. Treat stereo as gain, not as the only stacking method.
• Do not make critical targets findable only once they “float off the screen”; in a monocular screenshot the target should still be nameable from contour and layering.
• Allow stereo strength to be turned to zero, and keep the task completable at zero — that is accessibility, and also the path for temporary monocular use.
• How to check: run the full flow with one eye (or with identical left/right images) through grasping and panel ordering. Failures are where monocular fallback is missing. A stereo-normal developer’s self-test is not a sign-off.

Related

• Same group: N1.10.1 Binocular disparity only provides useful depth at near distances · N1.10.2 A wrong rendered IPD globally changes size judgments · N1.10.4 Reducing disparity eases the conflict at the cost of flattening space
• Nearby: N1.04 Vergence–Accommodation Conflict · N3.06 Transparency and Occlusion
• Search terms: stereoblindness · monocular depth cues · stereoanomaly