Distinguish "not done" from "not recognised"
Aliases: false negative · false positive · gesture segmentation · skeleton overlay
What it is
Motion-sensing failure has two sources: the player did not perform the target gesture (non-performance), or the player did but the sensor missed it (non-recognition). The fix paths are entirely different—the former is a teaching problem, the latter an engineering problem. When a game presents both under a single "not recognised" message, the player cannot tell whether to adjust the gesture or adjust the position, and the frustration comes from missing information rather than the failure itself.
Why it happens
Distinguishing them depends on making gesture segmentation and threshold decisions observable. Gesture segmentation cuts the continuous sensor stream into candidate action segments; only a successful cut enters the recognition comparison. A segmentation failure (noise, occlusion, dropped frames) means the gesture never even entered the recognition pipeline. Segmentation failure and threshold failure have different internal signal signatures: in the former, no candidate segment exists in the data; in the latter, a segment exists but its similarity score falls below threshold. The system can tell these apart internally, but most games never expose that internal state—both failures look identical on screen.
Where it stops holding
Perfect discrimination is not always feasible. The root cause of a segmentation failure (occlusion, lighting, excessive speed) may not be diagnosable in real time, and forced classification risks producing a wrong message. The value of distinguishing also depends on whether the player can act on it: if the message says "gesture too small" but the player believes it was big, the message itself creates a new attribution conflict—at that point auxiliary visualisation (skeleton overlay, amplitude bar) is needed to show the player what the system sees. Exposing distinctions also has a cost: another state indicator adds visual noise, and immersion-sensitive games must weigh the trade-off.
Applying it
- Preserve the segmentation-versus-threshold distinction inside the recognition pipeline and surface it as two distinct UI states: "no candidate gesture" versus "gesture similarity below threshold."
- Offer auxiliary visualisation (skeleton overlay, amplitude indicator) so players can compare what the system detects against what they feel themselves doing.
- Verification: deliberately produce both failure types (skip the gesture, perform it with occlusion) and check whether the UI message differentiates correctly. Then observe whether players take the repair action the message implies (adjust position vs adjust amplitude).