Presence is not the same as motion or a single detection
Motion detection says a region changed. Object detection says a person was found in a frame. Neither, on its own, answers the question that most real use cases actually care about: is a person actually present, sustained over time, in a way worth acting on? A single frame's detection could be noise; a flicker of motion could be a shadow. Presence is the layer that turns momentary signals into a confirmed, ongoing fact: someone is here.
What dwell time measures
Once presence is confirmed, dwell time is simply how long it lasts — how long a confirmed presence remains within a defined zone. It's a deceptively simple number that underlies a lot of practical use cases: how long a customer stands at a display, how long a vehicle waits at a gate, how long someone remains in a restricted area, how long a queue takes to clear.
Why presence is harder to get right than it sounds
A naive implementation — "count consecutive frames with a detection" — breaks in ways that are easy to predict once you've seen it happen:
- A person who stops moving can fade out of motion-based detection entirely (see Motion Detection's "stopped object" problem), even though they're still standing right there.
- Brief occlusion — someone stepping partly behind an obstacle — shouldn't reset a dwell timer to zero. The person didn't leave; the detector just briefly lost them.
- Per-frame detector flicker — a confidence score that dips below threshold for one frame and recovers the next — shouldn't be read as "the person left and came right back."
A reliable presence system has to reconcile all of this into one continuous signal, rather than treating every momentary drop in detection confidence as a real absence. That is precisely what makes presence a harder engineering problem than it first appears.
Presence without recognition
An important and often overlooked property: presence doesn't require knowing who someone is. A system can confirm "a person is here" and measure how long, using continuity and tracking, without any facial recognition or re-identification involved at all. That distinction matters both technically (it's a simpler, more robust problem than identity matching) and for deployment — many use cases genuinely only need count and duration, not identity, and building the system that way is both more private and less to get wrong.
From dwell time to loitering
Loitering detection is dwell time plus a policy: is this duration, in this zone, at this time, worth flagging? A person standing 30 seconds at a bus stop is unremarkable. The same 30 seconds in a restricted stairwell after hours might not be. The dwell measurement underneath is identical — what changes is the threshold and the zone-specific context applied on top of it. This separation — measurement as one layer, policy as another — is what makes the same dwell mechanism usable across very different alerting rules without rebuilding the underlying detection each time.
Where Vision Lab fits
Vision Lab treats presence and dwell as a first-class layer rather than an afterthought bolted onto detection, and runs it entirely on the edge so footage stays on site. It's the engineering foundation behind Vision Lab Studio, Vision Box, Cabin Cam, and Spy Catcher.
Frequently asked questions
What is presence detection?
Presence detection is confirming that a person (or other object of interest) is actually in a zone, sustained over time — distinct from a single flicker of motion or a one-frame detection that may or may not be real.
What is dwell time?
Dwell time is how long a confirmed presence remains within a zone. It's the basis for loitering detection, queue-length estimation, and any use case that cares about duration, not just a momentary appearance.
Why is presence harder than it sounds?
A person who stops moving can fade out of motion-based detection; a person who's briefly occluded shouldn't reset the dwell timer to zero; a detector's per-frame flicker shouldn't be read as the person leaving and re-entering repeatedly. Presence has to reconcile all of this into one continuous, trustworthy signal.
Does presence detection require identifying who the person is?
No. Presence detection confirms that someone is there and measures how long, without any facial recognition or re-identification. It answers "is someone here and for how long," not "who is this."
What is loitering detection built on?
Loitering detection is dwell time compared against a threshold for a specific zone and context — a person standing for 30 seconds at a bus stop is normal; the same duration in a restricted area at 2 a.m. may not be. The dwell measurement is the same mechanism; the policy layer on top decides what counts as noteworthy.
Why is a single detection not enough to raise a dwell alert?
Because one observation says nothing about duration. Dwell is a claim about time, so it has to rest on sustained, corroborated evidence gathered across many frames rather than on any single detection — which is also why dwell is a fundamentally different problem from detection.
← Back to Knowledge