Calibration makes the geometry correct. This section checks whether the installation behaves correctly for a person walking up to it.

Test order

Work through these in sequence — each one assumes the previous passed.

#TestPass condition
1Coverage sweepWalking the full active area produces a response everywhere, no dead zones
2Edge accuracyStanding at the four corners, the response lands under the person
3LatencyResponse arrives within the project’s budget (typically < 100 ms)
4Multi-targetTwo people tracked independently; no swapping of identities
5Exit behaviourLeaving the area releases the interaction cleanly, no stuck state
6Ambient lightWorks with house lights on, projection bright, and sunlight if applicable
7Power cycleEverything recovers without manual intervention

Coverage sweep

Walk a grid, not a line. A single pass through the middle will miss exactly the failures that matter — occlusion from a mounting bracket, a blind wedge behind the sensor, and the edges where the scan is least accurate.

Latency

Measure it rather than judging by feel: touch a known point, record the time between the physical event and the visual response. If it is over budget, check the audio/video pipeline before blaming the sensor — the sensor’s contribution is usually a few tens of milliseconds.

Common failures at this stage

SymptomLikely cause
Tracks in the centre, misses the edgesCalibration geometry or reference points
Works standing, fails walking fastFrame rate or host-side tracking, not range
Ghost touches with nobody thereReflective surface, or direct sunlight into the window
Interaction sticks after exitHost-side state machine, not the sensor

Handover checklist

  • Sensor address recorded in the installation table
  • Calibration profile saved and named after the site
  • All seven tests above pass
  • Someone other than the installer has walked the scene

Next

Something not passing? Go to Troubleshooting.