Offer interaction at a shorter distance than the sensor can still see

What it is

The farthest and nearest distances at which a sensor still reports points are larger than the distances at which gestures stay stable and people can still stand. Interaction distance should sit inside the capture range, with margin at both ends. On the far edge, skeleton jitter is already centimetres and fingertip pinch dies first. On the near edge the hand fills the frame, focus fails, and tracking drops too. Promising interaction out to the capture limit invites people to stand where signal-to-noise has already collapsed, then reads the failure as a bad gesture. How volumes nest, and how leaving the volume differs from a recognition miss, are finer spatial questions. Distance comes first: the stance you allow must be narrower than “still visible.”

Why it happens

Optical and depth resolution worsen with distance; noise grows with distance. Far away, the same joint angle occupies fewer pixels, and a swipe's span shrinks to the size of the jitter. Too close, occlusion, overexposure, and missed focus. People are also bounded by arm length: reaching the far edge straightens the arm, and fatigue and jitter rise together. “Still visible” is not “still a place to operate.” Set the published distance on the flat part of the capture curve, where the arm need not straighten, and failures at the edges will mostly be gesture failures rather than stance failures. Furniture cuts another slice: if sofa-to-television distance sits on the far edge, the product fails in the most common stance.

Studying it

Sweep radially. At fixed steps measure recognition, jitter (joint-position standard deviation), and reported fatigue. Draw three curves and take interaction distance as their intersection: recognition still high, jitter still below gesture amplitude, arm not straight. Measure the near edge too; do not only quote maximum working distance. Vendor maximum detection range is a capture limit, not a citation for interaction distance. Use the real seat depth of the target room, not a laboratory rail.

Where it stops holding

Headset near-field “distance” is tens of centimetres from eye to hand; margin lives more at the field-of-view edge than radially, but the rule holds: do not promise interaction in the last centimetres before tracking dies. Multi-camera stitching makes capture huge; interaction distance should still follow a comfortable radial for one-handed work, or people will stand on a seam. In the dark, infrared's far edge is nearer than daylight visible; set interaction distance for the worse light. Children cannot reach an adult's comfortable distance; the near edge is harsher for them.

Applying it

• Spec capture distance (“sensor still reports”) and interaction distance (“interaction allowed”) separately, with the latter strictly inside the former.
• Accept on the target room's seat depth, not on a laboratory maximum detection range.
• Do not blame the gesture first for far-edge failures; check whether joint jitter already exceeds movement amplitude.

Related

• Same group: C4.10.2 The coordinate anchor decides whether behaviour stays consistent after the user moves · C4.10.3 The boundary is invisible, so leaving it and a recognition miss look the same · C4.10.4 Boundary state must be expressed and distinguishable from recognition failure
• Adjacent: C4.20 Interaction volume versus capture range · C4.12 Fatigue cost of mid-air gestures
• Search: working distance · capture range · interaction distance