Weak texture and specular surfaces fail recognition

What it is

A freshly painted wall, a glass coffee table, a lit monitor — all look like “large planes,” and to visual SLAM they are often blank. Insufficient visual features keep plane detection from converging, or yield a fake plane that rides the highlight. The failure is optical, not a matter of the user missing the aim.

Weak texture lacks corners that can be locked again next frame. Specular surfaces add false features that travel with viewpoint. Both often live on the same piece of furniture.

Why it happens

A feature has to be recognized as the same point across neighboring frames. Flat paint, large solids, and repeating fine stripes offer nothing unique, so matching guesses or refuses. Specular reflection treats the image of a lamp or window as structure in the scene; when the head moves, that “structure” slides on the surface and the fitted plane twists with it. Time-of-flight and structured light drop returns on black, transparent, and mirror-like materials, so depth is hollow there too. An IMU can coast pose for a short stretch; coasting does not grow new planes.

So “this is a table” is not the same sentence for a person and for the tracker. People use silhouette and use; the tracker uses lockable texture. The most common placement targets in an office — white desks, glass, displays — sit in the tracker’s weakest set.

Studying it

Make surface material the independent variable: a wall with a poster, the same wall with the poster stripped, a glass table, a wood-grain table. Measure time to first plane, normal jitter once the plane exists, and the share of ARCore INSUFFICIENT_FEATURES or ARKit limited reasons that cite lack of features. Sweep lighting direction as well; specular failure moves with the lamp.

Dependent variables: detection latency, normal variance after the plane stabilizes, task abandonment. Do not measure only in front of a lab calibration board — the board exists to make features sufficient, so it cannot show this failure.

Where it stops holding

Devices with LiDAR can still produce geometry on some textureless surfaces, so weak texture is no longer equal to failure — glass and monitors often still are. Temporary texture (tape, fiducials) turns failure into success; that is an experimental intervention, not an environment the product can assume. In full dark the dominant cause is exposure, not texture. A repeating carpet can supply many features and still merge two different patches into one; that is a different matching error, not “no features.”

Applying it

• Write “white walls, glass, screens” as known failing surfaces. Steer placement toward textured regions (grain, seams, frames), not toward those.
• When a plane refuses to appear, make the prompt about the surface (“try a patterned tabletop”), not a generic “keep walking.”
• If a showroom or retail piece must sit on a glossy table, add trackable texture at install time rather than hoping the live tracker suddenly strengthens.
• How to check: seat the same virtual lamp on a wood-grain table and on a glass table. No plane, or a plane that follows the lamp, is this failure being hit. Do not accept only on a lab wall papered with posters.

Related

• Same group：N5.03.1 Placement depends on recognizing environment geometry · N5.03.3 Recognition failure needs explicit feedback and a manual fallback
• Nearby：N5.04 Spatial Anchors · N3.10 Depth Conflict and Clipping
• Search terms：insufficient visual features · textureless surface · specular tracking failure