AI Security converts existing CCTV footage into an auditable record of how working time was spent — per person, per hour. Time at the workstation, time on a phone, time away from the desk.
Submit a clip and AI Security returns identified people, detected objects, per-person timings and a written summary.
Register a face once and AI Security reports that person by name in every future scan, with a margin test so a near-miss stays unknown rather than becoming a wrong name.
Working, phone and away time per person, as numbers you can put in a report and trend across weeks rather than a feeling about how the day went.
Eighty object classes with unique-track counting, so a single chair observed for an hour is counted once, not ten thousand times.
Every scan closes with a short written read-out of what happened, so the outcome of a shift is legible without interpreting a table.
An annotated video comes back with every detection drawn on it, so any number on the dashboard can be checked against the footage that produced it.
Footage and face data never leave the cloud environment you already control, and sign-in is handled by your own identity provider.
Each frame a person appears in resolves to exactly one state, so the totals always reconcile against the time they were on camera.
Holding one position at a desk for a sustained stretch, with no phone in hand.
A phone resolved inside the person's own frame, counted wherever they are standing.
On camera but anchored nowhere — between desks, standing, passing through.
Absolute stillness over minutes, so a jacket left on a chair is never recorded as a person.
Most systems get this wrong in a way that quietly inverts the answer.
The intuitive rule is that a person overlapping a chair must be seated. On a real ceiling-mounted camera that collapses immediately, because the chair someone sits on is hidden behind their own desk. Tested against live footage, the overlap test held for none of the seated samples — while every empty chair across the room detected perfectly.
A system built that way reports the people who are working as absent, confidently.
AI Security decides a person is at a workstation when their position stays pinned to one spot for a sustained run. The reference point never drifts toward them, so walking can never be mistaken for settling in. It survives full occlusion of the chair, the desk and the lower body, and needs no per-camera configuration.
Where you want precision instead of inference, define explicit seat zones with
SEAT_ZONES and they take priority immediately.
Unrecognised faces are grouped by similarity, so eight sightings of one visitor count as one person, not eight — which is what makes "how many strangers came through" a number you can trust.
Every threshold is defined in seconds rather than frames. A fast scan that samples across a clip and a deep scan that reads every frame return the same verdict — confirmed by running both against identical footage and comparing totals.
Upload a clip, paste a link, or pick stored footage. Awkward codecs are converted without asking.
Long video is divided into segments, each dispatched as an independent job.
Segments process in parallel — detection, identity and state — streaming previews as they land.
Results merge, faces re-cluster across the whole video, and per-person hours are totalled.
A written summary, an annotated video, and a stored record you can trend over weeks.
No pilot programme, no installation and no change to how your teams work. A single clip is enough to evaluate the figures AI Security produces.
Get startedSign in to open the Camera Dashboard
Continue with MicrosoftSign in with your work or school account from any organisation.
Need access for your organisation?
Tell us who you are and we will provision your organisation
We reply within one business day.
Back to