Solutions
Start with the problem you need to solve first. Each workflow below says who it is for, what happens step by step, what it needs and where it stops.
- 1Shown on the video wall · 02:14 · zone entry at Gate 3
- 2Reviewed at a desk · Still frame and location checked
- 3Called in by radio · 02:16 · the operator calls a guard
The six workflows
- Incident investigationFind what was seen around an incident, and check it against its source.
- Video operationsCover more cameras with a queue of alerts that lead to their evidence where there is any.
- Vehicle searchFind a vehicle by plate, type and colour, and see where it was read.
- Gate and access correlationKeep a sighting of each arrival at a gate, and compare it with what your access records say.
- Occupancy and queuesCount people and vehicles by area and hour, and know when a threshold is passed.
- Site safetyBring hazard-zone entries and visible hazards to a reviewer, with a record of what was done.
For investigators who review footage by hand today.
Incident investigation
Start from a time and a place, or from an alert. Search the sightings, group the ones that look alike, and check each against its still and your recording before it goes into a file.
Watch an investigationBring in the footage
Connect the cameras, or upload the exported clips for the period. Uploaded clips are placed at the time they were recorded.
Search by description
Filter by object type, colour, plate, camera and time.
Group the candidates
Ask for look-alike sightings across cameras. A group is a candidate for the same person or vehicle, not an identification.
Check the source
Open the still for each sighting where one was captured. Replay uploaded footage in the product; for a live camera, go to your recorder.
Record the finding
Today a finding is recorded in the product only as a note on an alert. An investigation that starts from a search or from uploaded clips has no alert, so the finding goes in your own case file.
What it needs
- Footage for the period, as live cameras or exported clips
- A role that may search
Where it stops
- Grouping rests on overall appearance and can group similar-looking people
- A live sighting has a still, not a recording
- Review decisions on candidates are not stored yet
For control rooms with more cameras than people to watch them.
Video operations
Rules run on the cameras you connect. Operators watch a wall of the cameras that matter and work a queue in which each alert shows its rule and its status and leads to its still where it has one.
See alerts and responsePut the cameras on a wall
One to sixteen cameras with detections drawn on. Each feed says whether it is live, behind, stalled or lost.
Start from a rule set
Begin with a starter set, most of which depends on AI scene reading, then draw zones and set schedules, severities and repeat limits.
Add line and dwell rules
A line crossed in a chosen direction, or a stay in a zone beyond a set time. Verified by automated tests; not yet confirmed on a live feed in our own testing.
Work the queue
Each alert stays New until a named person marks it Acknowledged or False positive. Assign it, or record your assessment as a note on the alert.
Watch camera health
Cameras are shown as online, degraded or offline, and a lost feed raises its own alert.
Tune by the false-alarm rate
The rate by rule and camera shows what to adjust, and a rule reviewers keep marking false on a camera is muted there.
What it needs
- Live cameras the deployment can reach
- Operators with a role that may triage alerts
- Reviewers who mark alerts, so the false-alarm rate means something
Where it stops
- Acknowledging records who took an alert on; it is not a finding and does not close a case
- VisionObject runs on one machine: it should not be the only way an event is noticed
- Alerts that depend on an optional module pause when that module is paused
- Alerts are read in the dashboard and on the event stream; they are not sent by email or message
For teams asked "where did this vehicle go?"
Vehicle search
Plates are read where they are legible and kept with each sighting. Search by plate, exact or allowing for a misread character, and see the readable sightings as stops on a map.
How recognition worksRead plates where they are legible
On vehicles large enough in the frame, with a confidence value. A plate that cannot be resolved is skipped, not guessed.
Search exact or near
A near match differs by one character or a commonly confused one, and is flagged so a person can compare the two.
See the route
Stops in time order on a map. Lines between stops are inferred; the gaps are where no camera read the plate.
Search without a plate
Filter by vehicle type and colour, and group look-alike sightings as candidates.
Watch for a plate
A watchlist raises an alert when a listed plate is read. It has not yet been confirmed on a live feed in our own testing.
What it needs
- Number-plate module switched on
- Cameras placed so a plate covers enough of the image
- Camera locations entered by an operator, for the map
Where it stops
- Not every visible plate is readable: on night and dawn street footage only a small share of vehicles gave a usable read in our own testing, and daytime rates have not been measured
- A route with more than one stop has been verified by automated tests only
- A stop without a camera location is listed but not drawn
- A route is the set of places a plate was read, not the path the vehicle took
For sites where the gate log and the camera tell different stories.
Gate and access correlation
Each vehicle and person the gate camera follows becomes a sighting with its time and, where a frame was captured, a still. Comparing those sightings with visitor or door records is done by a person today.
See how communities use itKeep a sighting of each arrival
Each vehicle and person the gate camera follows becomes a sighting with its time and, where a frame was captured, a still.
Read plates at the gate
Where the camera and the light allow, the plate is kept with the sighting.
Add direction with a line rule
A line across the gate raises an event, with direction, when something crosses it. Crossings inside the rule's repeat limit are not recorded, so it is not a complete log.
Add the hours that matter
A zone at the gate, on a schedule, raises an alert on an entry during closed hours.
Compare with your records
A guard or reviewer checks the sighting against the visitor book, gate system or door log by hand.
What it needs
- A camera that sees the full width of each lane
- A zone or a line drawn on each gate camera, for the rules
Where it stops
- VisionObject does not know who was expected or approved
- A rule knows the time and the place, not the person
- A line rule is not a complete log of crossings: an event is recorded only when the rule fires
- A watchlist alerts on the plates you name; there is no alert on vehicles that are not on a list
For operations managers who need numbers on how spaces are used.
Occupancy and queues
Count rules and counts over time come from the same cameras your security team uses. Each figure is a count of the people or vehicles a camera is following. Each of them is also kept as a sighting, with a still where a frame was captured, for the retention period; there is no count-only mode.
See analyticsSet a threshold
An event when the number of people in the view of a camera that covers a queue area, a hall or a platform reaches the number you choose.
Read counts over time
People and vehicles by type, for each camera, by quarter-hour, hour or day.
Filter by area
Label cameras with their floor, zone or department and filter the figures by camera or area. The camera-health series over time covers the whole organisation and cannot be narrowed.
Compare with last period
Each headline figure is shown beside the previous period.
Review each threshold alert
A count rule that fires is an alert. It stays New until a named person marks it Acknowledged or False positive, so a threshold set too low shows up in the false-alarm rate.
What it needs
- A camera whose view is each area you want counted, without heavy occlusion
- Area labels on the cameras you want to filter by
Where it stops
- A count of people the camera can see, not a calibrated density
- A count covers a camera's whole view; it cannot be limited to a drawn zone
- A space covered by several cameras needs a rule on each
- A wait time is not estimated: the figure is a count of the people in view
- A people count is not a sales conversion rate
For safety teams who need context, not a replacement for safety systems.
Site safety
Zone rules cover marked hazard areas, and dwell rules cover bays and lanes. Each alert stays New until a named person marks it, and can be assigned and noted, so a weekly review starts from a list.
See how industrial sites use itMark the hazard zones
Draw each zone on the camera's view and set the shifts when entry should raise an event.
Set dwell limits
An event when a vehicle or person stays in a bay, lane or zone beyond a set time.
Review each event
A named person marks the alert Acknowledged or False positive, assigns it and records the follow-up as a note on the alert.
Add visible-hazard candidates
With AI scene reading, a person on the ground or visible fire or smoke is brought to a reviewer after the sighting.
Read the week
Counts by rule and camera, and the false-alarm rate, show which zones and shifts need attention.
What it needs
- Cameras that see each whole zone
- A named reviewer on each shift
Where it stops
- Not a safety function: it does not replace guarding, interlocks, alarms or supervision
- There is no guaranteed response time
- A scene reading follows the sighting and can be wrong
- Protective equipment such as helmets and vests is not checked
Questions we are asked first
Which workflow is quickest to start with?
Incident investigation on recorded footage. It needs no connection to live systems: you upload exported clips and search them.
Can security and operations share one deployment?
Yes. Both use the same cameras and the same record of sightings and events. A viewer cannot run plate, scene or route searches or group sightings, but can browse sightings, stills, events and alerts on the cameras in their scope. There is no figures-only role, and filtering sightings by colour is open to every role.
Does recorded footage raise alerts?
No. Uploaded footage is analysed for search and reporting only. Alerts come from live cameras, so nobody is alerted about something that happened last week.
What if we need something that is not here?
Tell us. A pilot is scoped on what is in the current release, and we say plainly where it stops rather than build a pilot around something that is not there.
Start with the workflow that costs you the most time.
Tell us whether it is incident review, alert noise, vehicle search or space planning. We will scope a pilot around that one problem.