Person Dwell Detection: Lingering & Long-Notice Alerts

Certain areas should never hold a person for long. An ATM lobby after closing, a cash-counting room, a server corridor, the stockroom of a retail store – anyone standing there longer than it takes to pass through is either waiting for an opportunity or casing the site. Person dwell detection closes exactly this gap on the cameras already installed: it measures how long someone remains inside a defined zone and alerts the moment they exceed the time you allow.

Person dwell detection is a timing rule layered on top of person detection. It is “how long has this person been here, and how many times have they come back”. That matters because the prelude to most internal theft is not a dramatic break-in; it is a person who lingers, leaves, and returns.

What the Rule Measures

Each sensitive area is drawn as a zone on the camera view. For every person the rule detects inside that zone, it maintains two counters: a dwell timer that accumulates how long they have stayed within the current visit, and a re-entry counter that tracks repeat visits within a window you set. A cleaner walking through in forty seconds never comes close to the threshold; someone standing at the stockroom shelf for six minutes does.

Both counters are configurable per zone, because “too long” means different things in different places. A customer at a self-checkout may legitimately stay three minutes; a person inside the cash office should be gone in one. When either counter breaks its limit – a single long stay, or repeated short visits that together look like casing behaviour – the rule raises an alert with the clip attached.

How It Runs on the Edge

The rule builds on object detection and runs on the edge appliance beside your existing cameras. Timing is the hard part of any dwell rule, and it is the same principle used across video analytics: a person briefly occluded by a shelf, a doorway pillar, or another colleague must not reset or break the timer. The tracker bridges short occlusions, so the clock keeps counting through moments when the person is momentarily out of view.

Because the analytics run locally, the rule keeps working through network outages – the condition under which someone testing your coverage is most likely to linger. Alerts queue on the appliance and deliver when the link returns, with the full dwell record intact for review.

How It Differs from Loitering Detection

The two rules are neighbours and cover different failure modes. Loitering detection watches a broader region for people wandering without purpose over longer windows – a parking area, a street frontage, a perimeter approach. Person dwell detection is tighter: a specific zone, a specific allowed duration, a counter that starts the moment someone enters. Use loitering when the question is “is someone hanging around the approach”; use dwell when the question is “why is anyone still inside this room”.

For group behaviour, crowd gathering detection flags when several people collect in one spot – a different signal from a single person lingering. And for after-hours presence, intrusion detection answers the harder question of whether anyone should be there at all. Sites commonly run dwell inside the building and intrusion and loitering on the approach, all on the same cameras through one appliance.

Where Sites Use It

Retail uses it on stockrooms and cash offices during trading hours, where dwell alerts pair naturally with the floor-level analytics described in AI cameras in retail stores. Banking and financial service points apply it to ATM vestibules and back offices. Warehouses use it on high-value cages and dispatch offices.

Care and supervised environments use the same timer differently: a person remaining too long in a bathroom or a resident’s room can be a welfare signal rather than a security one, which is where the rule connects to fall detection and the resident-side view in elderly fall detection. A dwell alert followed by a person no longer upright is the combination worth escalating immediately; the background on how that works is in what fall detection actually measures. Left-behind items that often mark the spot where someone lingered are covered separately by abandoned object detection.

For the wider site design around this, start from apartment complex security cameras.

For the background on how this works, see virtual fence system.

See it applied end to end in our guide to data center security cameras.

For the wider site design around this, start from church security cameras.

See it applied end to end in our guide to presence detection on camera.

Frequently Asked Questions

Isn’t this the same as loitering detection?

No, and the difference matters for tuning. Loitering detection looks for aimless presence across a wide region and typically uses longer windows; person dwell detection enforces a maximum stay inside a specific zone and can fire in under a minute. Many sites run both, but the sensitive-zone rule should be dwell – it is precise, and precision keeps operators trusting the alerts.

Can each zone have its own dwell time?

Yes. Every zone carries its own threshold, re-entry window, and active schedule. A cash office at one minute and a stockroom at five minutes coexist on the same camera or the same site, and each policy only applies during the hours you assign to it.

How long should the threshold be set?

Start from the longest legitimate transaction you expect in that zone, then add margin – for example, double the normal duration of a cash-office task. Run the rule in review mode for a few days, look at the dwell distribution it records, and set the threshold just above the legitimate tail. The recorded timeline makes this an evidence-based setting rather than a guess.

What stops someone resetting the timer by stepping out?

The re-entry counter. Short exits and returns inside the window you define do not clear the record; the rule can alert on repeat visits even when each single visit stays under the threshold. Casing behaviour rarely survives that setting.

Next Step

Send us the zones you want timed and the maximum stay each one allows. From the 198+ algorithm library this is configuration, not development – the first demo takes about 7 days where the detection already exists in the library. Start with a Starter Kit at USD 999, or tell us your scenario for a custom build quoted per site.