Head motion, glasses slippage, and posture change all worsen drift

What it is

A common mechanical source of drift is relative motion that keeps happening in a session: head versus camera, glasses versus the nose, trunk versus the chair. On a wearable, loosening straps let the cameras crawl on the face; on a remote tracker, a shift in the chair walks the head out of the calibration box. The motion need not be large enough for the user to count it as “I changed pose”; millimeters suffice to push the mapping off. This explains why drift speeds up; it does not define what drift is, and it does not decide when a one-shot condition change voids calibration.

Why it happens

Head-worn cameras assume rigidity between camera and eyeball. Strap elasticity, sweat, and zygomatic motion while talking break that rigidity; the cameras’ extrinsics relative to the corneal glint change. Glasses are a second sliding rigid body: the frame drops on the pads, prism and reflections move together, and the corneal hotspot jumps to another point on the lens. Posture change slowly alters head height and distance through spine and hip, and a remote camera’s perspective follows.

The sources stack. During a film people lean in without noticing, push their glasses, and sweat under the headset brim; three curves climb at once. A laboratory chin rest and tape on the glasses turns the main sources off, and measured drift looks unrealistically pretty.

Studying it

In a long session in a natural sit, log in parallel: headset-to-face IMU or visual odometry, frame position relative to the inner canthus if available, chair-back angle, and gaze validation error. Correlate, or intervene with a small slip (“please push your glasses but do not calibrate”). Separate “the user thinks they did not move” from instrument-measured millimeters. Makeup, a headscarf, and a mask change strap and glasses friction and must be strata, not noise to delete.

Where it stops holding

A custom 3D-printed face rest, firm straps, and no glasses can make the slip arm weak, and drift is then dominated by pupil and tear film. A fully rigid bite-bar research mount almost kills this path and does not represent a consumer headset. Conversely, running, a workout, or a bumpy car turns “worsen” into the dominant term, and drift is fast enough to need continuous compensation. After unilateral temple pain or dental work, people don the headset crooked and slip is there from the first second.

Applying it

• In headset copy, treat strap tension and glasses position as things not to change after calibration; offer a lightweight remeasure rather than assuming no slip.
• When a clear headset-on-face slip or a jumped glasses hotspot is detected, raise a drift warning early instead of waiting until people miss.
• Verify error climb in sessions with sweat, speech, and natural glasses-pushing, against a chin-rest session, confirming that product numbers were not taken on a rigid mount.

Related

• Same group: C8.10.1 Drift is the gradual offset of gaze estimates from true position after calibration · C8.10.3 Implicit recalibration can correct drift from interaction with known targets without interrupting the task · C8.10.4 Beyond a point, implicit correction is not enough and the user must actively recalibrate
• Adjacent: C8.08 Eye-tracking calibration · C8.05 Head pointing
• Search: slippage · headset fit · glasses