Connect
Link sightings across cameras, read a scene in words, replay uploaded footage as one timeline with the gaps between clips skipped, and summarise a week. Each analytic below states what it measures, what it must not be used for, and how far it has been evaluated.
What is included
Analytics that go beyond a single camera or a single frame, each with what it measures and how far it has been evaluated.
From the pictures alone
Work from images, with no other system involved.
Grouping across cameras
Optional moduleCandidates for the same person or vehicle, grouped by overall appearance.
Limit: Groups rest on overall appearance, so similar clothing or similar vehicles can be grouped together.
AI scene reading
Optional moduleUses an external AI serviceA short description, up to eight tags and a suggested severity for a sighting.
Limit: A reading arrives after the sighting; it is not on the real-time alert path.
Describe-the-scene search
Optional moduleFind sightings by what their scene reading says.
Limit: Only sightings that received a reading can be found this way.
Continuous playback
Uploaded footage replayed as one timeline, with the gaps between clips skipped.
Limit: Covers uploaded footage only; a live sighting has a still, not a recording.
Weekly brief
Optional moduleUses an external AI serviceA weekly fact sheet with a written summary.
Limit: The figures are computed first and kept; if the summary cannot be written, the brief says so and shows the figures alone.
Plates across cameras
Verified by automated tests; not yet confirmed on a live feed in our own testing.
Vehicle route by plate
Optional moduleOne plate's readable sightings as a time-ordered route on a map.
Limit: Verified by automated tests; the only route recorded on live footage in our own testing had one stop.
Plate watchlist alerts
Optional moduleAn alert when a watched plate is read.
Limit: Verified by automated tests; not yet confirmed on a live feed in our own testing.
Analytic cards
What each model-based analytic measures, and how far to trust it.
For each one: the signal, what it is for, what it is not for, how it has been evaluated and how a person reviews it.
Grouping across cameras
- Input
- Video
- What it measures
- How alike two sightings look overall: clothing, colour, shape. Compared within the same object type only.
- Intended use
- Narrowing a search to sightings worth a reviewer's time.
- Not for
- Identifying a person, or deciding that two sightings are the same person without review.
- Evaluation
- Tuned on our own test footage from one camera; matching across cameras has not been measured. No published accuracy figure; precision is measured on a reviewed sample of your footage in a pilot.
- Review
- A reviewer compares the stills. Similar-looking people are grouped together at times, and a run covers the 500 most recent sightings in view.
Number-plate reading
- Input
- Video
- What it measures
- Plate text and a confidence value, voted across the frames of one sighting.
- Intended use
- Searching for a vehicle and tracing its readable sightings.
- Not for
- Automatic enforcement or stopping a vehicle on a single unreviewed read.
- Evaluation
- Read rate depends on plate size, angle and light. In our own testing, on night and dawn street footage, only a small share of vehicles produced a usable read; daytime rates have not been measured. Measured per camera, day and night, in a pilot.
- Review
- The plate as read and the watched plate are shown side by side, with an exact or near-match flag.
AI scene reading
- Input
- Still frame
- What it measures
- A general vision model's description of one still frame: a summary, up to eight tags from a preferred list that the model may add to, and a suggested severity.
- Intended use
- Bringing a sighting to a reviewer's attention and making sightings searchable in words.
- Not for
- Real-time alarms, proof that something happened, or any conclusion about a person's intent.
- Evaluation
- No published accuracy figure. The reading instructions and the tag list are written for city street scenes in India. Agreement with your reviewers is measured tag by tag in a pilot.
- Review
- The still is shown beside the reading. A reading can be wrong or missing.
Events it produces
What this capability watches for.
Each event says what it needs and where it stops. The rules you set on a camera raise these events, and the alerts that go with them.
Rider without helmet
A scene reading describes a two-wheeler rider or passenger with no helmet.
- When
- The reading of a two-wheeler sighting carries the no-helmet tag.
- Limit
- Caps, hoods and distant riders are read wrongly at times. Treat it as a candidate for review, not a finding.
Triple riding
A scene reading describes three or more people on one two-wheeler.
- When
- The reading of a two-wheeler sighting carries the triple-riding tag.
- Limit
- Two riders close to a pedestrian, or two motorcycles side by side, can be read as three.
Person on the ground
A scene reading describes a person lying or collapsed on the ground.
- When
- The reading of a sighting carries the person-on-ground tag.
- Limit
- It follows the reading, so it is not an immediate fall alarm. People sitting or resting can be read as on the ground.
Weapon visible
A scene reading describes a visible weapon.
- When
- The reading of a sighting carries the weapon-visible tag.
- Limit
- It follows the reading, so it is not a real-time weapon detector. Tools, umbrellas and phones can be read wrongly; concealed items are not seen.
Fire or smoke visible
A scene reading describes visible fire or smoke.
- When
- The reading of a sighting carries the fire-or-smoke tag.
- Limit
- A reading is made when a person or vehicle is sighted, so an empty scene is not checked. It does not replace a fire detection system.
Outcome
The platform points to where to look. A person decides what it means.
Where it is used
See it working on footage like yours.
Tell us the workflow you care about. We will walk through it on demonstration footage, say what is available today, and plan a pilot on your own cameras.