Motion inside capture range but outside the interaction volume is easily mistaken for system failure

What it is

The hand is already waving where the camera can see it; a skeleton may even be drawn in a corner of the screen; the command does not fire. That location is inside capture range and outside the interaction volume. Users see “it is tracking me” and explain silence as a crash, a bad recognizer, or a wrong gesture, not as “I am standing outside the policy box.” When the two volumes disagree, the damaging case is not invisibility. It is being seen and not counting.

Why it happens

Debug overlays, mirrored preview, and joints are products of the capture pipeline and are usually drawn before the containment test. Users take those pixels as “the system is listening.” If rejection by the interaction volume has no separate account, the only behavior is silence—the same silence as recognition failure, not being engaged, or low confidence. The typical repair order is: repeat a bigger motion → switch hands → declare the device broken. Bigger motions are more likely to throw joints out of the volume or introduce self-occlusion, so failure reinforces itself. A skeleton drawn on an out-of-volume hand makes the contradiction sharper: the system “saw” it.

Studying it

Have participants perform the target gesture from a marked point that is inside capture range but outside the interaction volume, and note whether on-screen tracking visualization is present. Record their spoken attribution (broken / wrong gesture / wrong stance / not activated) and the strategy that follows (bigger motion, step closer, give up). Compare three feedbacks: total silence, tracking still shown but marked out-of-volume, and an explicit “please enter the operating region.” Dependent measures are the rate of correct attribution and time to reach the interaction volume. Do not test after telling them where the box is; that measures compliance, not spontaneous explanation.

Where it stops holding

Products with no tracking visualization hide “being seen,” so this contradiction weakens and silence is harder to tell from “the camera is off”—the problem shifts to whether capture state is visible. If the interaction volume almost coincides with capture range, the out-of-volume band is thin and this error shape is rare. Game players who already stand on a marked mat attribute faster to stance; first-time public-display visitors have no such convention. Child height slices the interaction volume differently relative to capture range; adult stance findings do not transfer.

Applying it

• When a hand is tracked but outside the interaction volume, give an account distinct from recognition failure—“I can see you; step forward into the operating region”—rather than letting the skeleton keep blinking.
• Hide user-facing debug skeletons, or draw a valid cursor only inside the volume, so out-of-volume tracking does not become false evidence of “it is listening.”
• At acceptance, start gestures from the out-of-volume capture band and count how many people report “it’s broken.” That rate is the user cost of mismatched volumes, not the recognizer’s accuracy.

Related

• Same group: C4.20.1 Capture range is the volume in which the sensor can physically detect motion · C4.20.2 The interaction volume is the subspace of capture range treated as valid input · C4.20.4 The interaction volume is typically smaller than capture range to leave a margin for tracking jitter
• Adjacent: C4.22 Perceiving interaction-volume boundaries · C4.29 Visibility of capture state
• Search: out-of-volume · tracked but ignored · false system failure