Short answer: Test the room with the number and movement patterns it is intended to serve. A single performer waving at a sensor proves only that one gesture can trigger a response in that condition. The opening test needs several people crossing, shielding one another, entering late and leaving, plus staff moving at the edge of the zone.
Ultratec describes a 12 m × 6 m multi-person sensor installation at ICOM 2025. Its case page establishes the supplier's stated design and room size, but provides no measured tracking-error rate. Research shows why the missing test matters. A 2015 MuseumVisitors dataset paper was explicitly built to examine group detection and occlusion in a museum. A 2016 interactive-wall study reports that its depth-camera tracker could lose a person's ID when someone was hidden or left the scene; the researchers tested their re-identification design with data involving 81 people. These are different systems and settings. Neither paper gives an acceptable error threshold for a new event room.
Name the response that the visitor should see
Start with a map of zones and cues. A useful specification says: “When a visitor enters zone A, the story acknowledges that zone; when two visitors move independently, one person's action does not steal the other's response; when tracking is uncertain, the room returns to a neutral state.” It should also define whether the room responds to individual positions, group presence or a single combined movement. If the creative concept only needs a group-level reaction, persistent identity tracking may be unnecessary.
Archv's group turn-taking guide covers who gets a turn at a shared experience. This test is different: it checks whether the sensing system correctly sees the intended presence and movement before the turn rules can work.
Rehearse movement, not just gestures
Use visitors or testers unfamiliar with the choreography. Test in the final lighting, audio and projection conditions, with the actual sensor mounts and furniture. Record the software version, sensor positions and number of people so a later fix can be retested fairly.
| Scenario | Observe and record |
|---|---|
| One person enters, pauses, leaves | Did the correct zone react and reset? |
| Two people cross paths | Did the response switch people, duplicate, or stay attached as intended? |
| A group briefly hides one member | Did the room keep a coherent state or trigger a misleading new story? |
| A late visitor enters during a cue | Could they join without restarting everyone? |
| A host moves outside the marked play zone | Did background movement trigger the public experience? |
| People cluster at the boundary | Did the room admit, ignore or flicker between states in a predictable way? |
This is a proposed acceptance script, not an outcome reported by the cited studies. Mark each attempt as correct response, missed response, wrong-person/zone response, false trigger or uncertain response. A raw count is useful only with the number of opportunities and conditions. “Five errors” in five trials means something different from five in five hundred.
Set a behavior for uncertainty
A sensor system will sometimes have ambiguous input. Decide what the public display does then: hold the current state briefly, show a neutral animation, ask for a clearer action, or let the host restart. The correct choice depends on the narrative and whether a wrong response would misstate content. Do not make the room confidently announce that one named visitor performed an action if the system only has uncertain movement data.
If the concept uses camera or depth data, document what is captured, retained and visible to operators, then ask the buyer's privacy owner to approve the actual configuration. The cited tracking research demonstrates technical possibilities and limits; it does not supply legal permission for an event deployment.
Sign off the busy-room state
Run the same script at normal and expected peak occupancy. Record failures by scenario, fix the cause and repeat the affected test. Include the host's recovery instruction and a reduced mode for a sensor outage. A system that works only when the supplier demonstrates it in an empty room is not ready for the group experience the buyer commissioned.