Binocular disparity only provides useful depth at near distances

What it is

The horizontal difference between where a point lands in the two eyes is binocular disparity. It shrinks fast with distance: the baseline is IPD, and once depth is large the visual axes are nearly parallel, disparity drops below the stereoacuity threshold, and stereopsis stops saying “how much farther”. At a desk, within an arm, disparity is a strong cue. At the end of a corridor, on a building outside, depth is carried by occlusion, perspective and motion parallax, not by the interocular difference.

The claim is about the working range of disparity as a depth channel — not about why vergence and accommodation fight, and not about how far a panel should sit for comfort.

Why it happens

Under the small-angle approximation, disparity is roughly IPD over distance. Double the distance, halve the disparity. The smallest disparity people can resolve is a few to a few tens of arcseconds, which, converted to distance, already swallows neighbouring depths beyond roughly ten metres. Headset pixels are coarser than lab stereograms, so the useful range is often nearer still — the screen-door pattern quantises fine disparity away.

Near space is the opposite: the same depth increment yields a large disparity, stereo matching is stable, and reaching to grab is mostly this cue. A distant object can still be rendered as two “correct” images; the retinal difference is already gone, and what remains is two slightly different flat pictures.

So “the whole scene is stereo” is a rendering fact, not a perceptual one. Budget depth by distance: only the near field should hand structure to disparity.

Studying it

The classical probe is a Howard–Dolman style depth discrimination: two rods, vary their relative distance, and watch threshold as a function of viewing distance. Random-dot stereograms (Julesz) strip monocular contours so the collapse of the disparity channel can be seen on its own.

Independent variables: viewing distance, IPD, relative target depth, pixel angular resolution. Dependent variables: stereoacuity, depth-discrimination threshold, farthest distance at which ordering still survives after motion parallax is removed.

In a headset add a control: head held still versus micro-movement allowed. If far depth suddenly improves with movement, motion parallax was carrying the scene, not binocular stereo.

Where it stops holding

A larger IPD yields more disparity at the same distance, so the useful range stretches a little; children and people with a small IPD lose the stereo difference sooner. Anyone using one eye, or whose stereopsis never developed, never had this distance curve. Optical see-through stacks real far scenery — itself almost disparity-free — against virtual near objects; the fight then sits in other cues. High-contrast thin rods in the lab are easier to “still see in stereo” than large colour blocks on a panel — shipping lab thresholds as product distances is optimistic.

Applying it

• Put structure that must be reached or ordered in depth inside a metre to a few metres, where disparity still has numbers to report.
• Do not separate skyboxes, horizons and far architecture by “more stereo”; use occlusion, aerial perspective, contact shadows.
• Make far labels and navigation arrows readable from monocular size and contrast. Do not build them as hairlines that only float once stereo is on.
• How to check: clamp the head, freeze motion parallax, and ask for a depth order on a near set and a far set. The near set should still sort; the far set should start to be guessed. If the far set also needs stereo to sort, pull the structure closer or add monocular cues.

Related

• Same group: N1.10.2 A wrong rendered IPD globally changes size judgments · N1.10.3 Some people have no stereo vision and need monocular depth-cue fallback · N1.10.4 Reducing disparity eases the conflict at the cost of flattening space
• Nearby: N1.04 Vergence–Accommodation Conflict · N3.02 Interface Distance
• Search terms: binocular disparity · stereoacuity · depth cue