Performance Hub LogoPerformance Hub

Troubleshooting

Symptom-first diagnosis for the people/AI layer. For camera, recording, upload, or appliance problems, start with CCTV Troubleshooting — everything here assumes video itself is healthy.

Nothing appears on the People timeline

The timeline shows "No face detections — There are no face detection items in this current view range of the timeline."

  1. Is People Tracking on? It's per-camera and off by default — check each camera in Cameras & devices.
  2. Is the detection-threshold filter hiding everything? By default people need enough detections over enough minutes to appear. Toggle the threshold filter off — if people appear, they were just below the minimums.
  3. Check the date window — the 1D/2D/3D range and date picker are easy to leave on the wrong day.
  4. Is the camera actually seeing faces? Person detection needs faces of usable size and angle. A ceiling camera looking straight down detects heads, not faces. Check the camera's framing in live view.
  5. Feature enablement — if the whole People area is missing or APIs return "People feature is not enabled for this facility", the feature isn't enabled; contact your account manager.

A member is never identified (always "Unsure")

The most common issue, and it's nearly always the photo:

  1. Check photo status. In the People list, find the member. An "Invalid Photo" badge (tooltip shows the score and fail reason) or "Not Processed" status means face matching has nothing to work with. Fix the photo in your member system, or edit it directly ("Edit Profile" → "Change").
  2. No photo at all? Self Learning can still identify them from visit correlation — check whether they're sitting in its queue, and confirm Enhanced AI Analysis is on.
  3. Identify manually once. The "Identify This Person" action on their unsure profile immediately links it, and their accumulating shadow photos improve future matching.
  4. Persistent near-misses (hover popover shows similarity just under 85%) → see Face recognition tuning — but improve the photo before touching the thresholds.

The same person keeps getting new "Unsure" profiles

Recognition isn't linking their visits together:

  1. Merge the duplicates — select the profiles on the timeline and merge. Every merge consolidates face references and makes the next visit likelier to link.
  2. Check camera image quality — dirty domes, backlighting (a camera facing a bright entrance), and low light all produce reference photos too poor to match against. The camera view is the fix, not the settings.
  3. If it's facility-wide, the shadow-profile quality preset may be too strict for your lighting — see the symptom table.

Wrong person identified

Rare, and worth treating seriously:

  1. Fix the record: open the person's profile and check their shadow photos — a mis-merge or bad auto-added photo is usually visible. Remove the wrong imagery, or merge/re-identify to the correct member.
  2. If Self Learning made the link, use "Unlink" on the record ("the self-learn record will be reverted") and "Reject" the match with a reason so it isn't re-proposed.
  3. Tighten matching: move Member Matching towards Strict — see Face recognition tuning.

A poster / mannequin / TV keeps appearing as a person

Working as designed, and there's a purpose-built fix: blacklist it. Open the false person's menu → "Add to Blacklist" with a reason ("e.g. Store Mannequin"). Future detections are dropped before any processing. If detections persist from a second camera angle, add a second blacklist entry from one of those detections.

Reports are empty or missing zones

  1. Reports lag by design — snapshots generate overnight for the previous day. A brand-new facility sees data from day two.
  2. "No zones configured — assign cameras to access zones in Settings" — zone analytics need access zones with cameras assigned.
  3. KPIs look low compared to check-ins — check which cameras have People Tracking on; report data only comes from analysed cameras.

Access violations look wrong

  1. False violations after setup are almost always rule bugs — a legitimate plan missing from a zone's allowed list. Review the zone's Membership Plans rules before assuming member behaviour.
  2. Remember the definition: a violation is a plan-rule mismatch — "excluded, or not in the allowed plans" — including time windows from Access Hours. A member on the right plan at the wrong time still violates.
  3. Open zones never raise violations — if you expect violations from a zone badged "Anyone can access", it needs rules first.
  1. Enhanced AI Analysis must be on — with it off, correlation doesn't run at all (and the settings section is hidden).
  2. Check the evidence bar — profiles need the minimum correlated visits (default 8). The In Pipeline tab shows progress per profile ("N/M visits"); infrequent visitors take weeks to accumulate evidence.
  3. Check-ins must be syncing — no check-in data, no correlation. Verify with the insights_checkins MCP tool or your Partner integration status.
  4. A big "No Match" tab is often couples/pairs who always arrive together — the system refuses to guess between two candidates. Manually identify one of them.
  5. A big "Unable to Link" tab points at member-system data quality — typically records missing the fields needed to complete the link.

A merge, identify, or delete seems stuck

These run as tracked background processes — re-attributing a long detection history takes time. Cards show transient states ("PROCESSING", "DELETING..."); merges show "Processing N of M…" with a "Retry Failed" option on partial failure, and failed deletes show "DELETE FAILED - RETRY". Retry once before contacting support with the person ID ("Copy Person ID").

When you contact support

Include: the facility name, the Person ID (copyable from any tile menu or profile), the camera involved, the date/time window, and — for identification issues — whether the member's photo status is valid. Anonymised screenshots of the hover popover (which shows the similarity score) shortcut most matching investigations.