Video intelligence for security and operations
Turn camera footage into a connected investigation.
VisionObject turns live streams and recorded footage into searchable sightings of people and vehicles. Review possible links between sightings, check each one against its source, and give security and operations teams the context to investigate and respond.
- 1Main gate · 22:41 · plate read
- 2Campus road · 22:43 · candidate, no plate
- 3Loading bay · 22:46 · yard closed
- 4Waiting for a reviewer
One alert, start to finish
Camera footage is the source. Context is the product.
Start with one sighting. Narrow the time and the cameras, compare the candidates, and check the source behind every link you rely on. Thirty seconds, drawn, with demonstration data.
One alert, in 30 seconds.
From a first sighting at the gate to a recorded response, in four chapters.
Read what the film shows
- See the sighting.
A drawn plan of a working site: a warehouse, offices, a car park, a perimeter fence and a public road. Cameras watch the gate, the roads and the loading bay. After hours, a vehicle drives along the approach road, turns in and stops at the main gate. A violet marker frames it and a label records the sighting: main gate, 22:41. The sighting is kept with a still frame. Live video stays on the site's own recorder.
- Link the candidates.
The view widens to the whole site. A violet thread runs back from the gate to an earlier sighting on the approach road at 22:38. The barrier lifts and the vehicle drives on; the thread follows it past the campus road camera at 22:43 and on to the loading bay at 22:46. An operator searches the plate and asks for look-alike sightings, and the sightings on the other cameras come back as candidates, in time order. Each is a candidate for a person to review, not a confirmed match.
- Check the source.
The thread leaves the site plan and becomes the spine of a timeline read beside one alert: vehicle in the yard after hours. Its status is New. The four sightings are listed in order with their times and places. An operator opens the stored still from the main gate, taken at 22:41, and checks the alert against it. The operator acknowledges the alert at 22:47, and its status changes to Acknowledged.
- Record the response.
The thread carries on and splits in two. One branch reaches the control room, where the alert has been on the wall display since it was raised at 22:46. The other reaches the office of the shift supervisor, where the alert, assigned to the supervisor at 22:47, appears on the desk screen. The supervisor adds a note at 22:48. The view pulls back to show the whole path, from first sighting to recorded response: seen 22:41, acknowledged 22:47, note added 22:48.
Illustrative workflow using demonstration data. Available features and integrations are confirmed during pilot planning.
Open the film as a video fileWhat each chapter stands on
See the sighting.
- A rule on a zone raises an alert from a live camera
- The clearest frame of the sighting is kept as a still
Link the candidates.
- Search by type, colour, plate, camera and time
- Look-alike sightings grouped on request, as candidates
- A plate's sightings as a route across cameras
Check the source.
- An alert opens with its notes and, where one was captured, the still of its sighting
- Footage you upload plays back with detections drawn on
Record the response.
- Acknowledged or False positive, under a person's name
- Every change written to the audit log
Describe and search
Every sighting is described. Most of the description can be searched.
Nine object types, twelve colours, what a person carries, what kind of vehicle it is, the plate where it can be read, and a reading of the scene. Pick one to see what it finds in the same street.
Object type
Nine types are detected in each analysed frame and followed within one camera.
Only these nine types are followed. Anything else in view, such as an auto-rickshaw or a hand-cart, is named only by a scene reading.
- Cyclist: Bicycle
- Motorcycle carrying three people: Motorcycle
- White car: Car
- Blue van: Car
- Orange truck: Truck
- Green bus: Bus
- Person in red with a backpack, at the main gate: Person, Backpack
- Person in yellow with a handbag: Person, Handbag
- Person in uniform carrying a box: Person
- Person under an umbrella: Person
- Suitcase by the bench: Suitcase
- Person in red with a backpack, on the campus road: Person, Backpack
- Person beside the hand-cart: Person
People
- Person
Vehicles
- Car
- Motorcycle
- Bicycle
- Bus
- Truck
Carried items
- Backpack
- Handbag
- Suitcase
The whole list, in words
- Source
Search within: One camera; Every camera in your scope; Uploaded recordings; A department or zone, for events and figures
- Time range
Kept for each sighting: First seen; Last seen; The clearest frame, as a still, where one was captured
- Object type
People: Person
Vehicles: Car; Motorcycle; Bicycle; Bus; Truck
Carried items: Backpack; Handbag; Suitcase
- Colour
Recorded as: People: upper and lower clothing; Cars, motorcycles, buses and trucks: main colour
The twelve names: black, white, grey, red, orange, yellow, green, cyan, blue, purple, pink, brown
- Worn or carriedShown, not searchable
Tags: Backpack; Handbag; Hat; Face mask; Uniform; Umbrella; On a phone; Carrying a box; Rider with helmet
- Vehicle
Type: Car; Motorcycle; Bicycle; Bus; Truck
Body style: Sedan; SUV; Pickup truck; Delivery vanShown, not searchable
- Number plate
Kept for each read: Plate text as read; Confidence of the read; Whether it fits a valid format
Search: The whole plate, or part of it
Route: Near match: one character out, or a commonly confused character; The cameras where it was read, in time order
- Scene reading
Present: auto-rickshaw; two-wheeler; pillion-rider; car; truck; bus; bicycle; hand-cart; pedestrian; street-vendor; stray-animal
Scene: heavy-traffic; light-traffic; junction; footpath; divided-carriageway; flyover; street-lit; night-scene; daytime; low-visibility; heavy-rain; camera-obscured
Vehicle: vehicle-wrong-way; vehicle-parked-illegally; vehicle-stopped-in-junction; vehicle-overloaded; vehicle-damaged; vehicle-no-plate; vehicle-plate-obscured
Two-wheeler: motorcycle-no-helmet; motorcycle-triple-riding; motorcycle-on-footpath
Person: person-on-roadway; person-running; person-loitering; person-on-ground; person-climbing; person-carrying-large-object
Crowd: crowd-forming; crowd-blocking-road; crowd-dense
Hazard: collision; fire-or-smoke; weapon-visible; unattended-baggage; road-obstruction; spilled-load; flooding
- Movement
Events: Entered a zone; Item left still beyond a set time; At least a set number in view
Also: Left a zone; Crossed a line, with direction; Stayed in a zone beyond a set time
- Similar appearance
A group gives you: Every sighting in it, in time order; The cameras it appeared on; The stills to compare
Not recorded
- Speed
- Vehicle make and model
- Object size
- Distance between people
Not classified by any model
- Emotion or mood
- Intent, dangerousness or likelihood of offending
- Gender, ethnicity or other demographic classification
- Who a person is, from appearance alone
An AI scene reading is instructed to describe clothing, vehicles and behaviour only. Its text is not filtered.
Workspaces
One record of what was seen, four ways to work with it.
Each workspace is built for one job. These are illustrative screens with sample data, drawn from the same evening the film follows.
Watch the cameras that matter, with their status in words.
One to sixteen cameras on a wall with detections drawn on. Each feed says whether it is live, behind, stalled or lost, so a stalled or lost feed is not read as a quiet scene.
What it does not do
VisionObject does not record live video. Your recorder does; the platform keeps stills of what it sees.
Industries
The same platform, set up for the place you protect.
Each market starts from a different problem, a different reviewer and a different set of measures. These six have a full workflow page; eight more are listed in the industry directory.
Public Safety
Investigation and response
Communities
Arrivals and shared spaces
Campus & Education
Gates, hostels and shared spaces
Industrial
People, vehicles and hazard zones
Airports
Perimeter incidents and terminal flow
Smart Cities
Shared cameras, separate responsibilities
Compatibility and deployment
Keep the cameras. Validate the connection.
We confirm source compatibility, operating conditions and the right processing profile before a pilot begins. VisionObject reads your streams alongside your recorder and runs in your environment.
- 1Cameras you already own · Building and mast
- 2Recorded footage · Exports from your recorder
- 3Other footage, as files · Needs assessment first
- 4One place to search · Streams and uploads together
- Streams, not replacements
- Live streams from IP cameras, recorders and video management systems are read as streams. Your recorder keeps recording; nothing is installed on the cameras.
- Stream, file or adapter
- Reading a stream and importing a file are supported today. A native adapter that drives a recorder's own playback and export is not offered.
- Your hardware or your cloud
- A workstation for a pilot is what we have run end to end ourselves. A server on site or a virtual machine in your own cloud account requires assessment first. It runs on one machine; high availability has not been validated.
- Sized before it starts
- Camera count, resolution, frame rate and the modules you switch on decide the machine. We size it with you.
Evaluation
Demonstrate the outcome on your operating conditions.
Agree a workflow, representative footage, evaluation measures and review responsibilities. Expand only after the pilot establishes what works and where the limits are.
Read the pilot guideWhat a pilot measures
Pilot scorecard
Agreed before anything is installed
- False-alarm rate
- From the alert queue, once your reviewers have marked the alerts.
- From your pilot
- Plate read rate
- From plate statistics, checked against a hand-labelled sample.
- From your pilot
- Candidate precision
- A reviewer labels a sample of groups against their stills.
- From your pilot
- Missed events
- Compare a hand-kept list for the pilot period with the alerts raised.
- From your pilot
- Time to acknowledge
- From alert statistics.
- From your pilot
What our evidence shows today
On one outdoor camera recorded over a single day, 53 clips were placed on one timeline and 116 sightings were grouped into 27 candidates for review. That is our own test footage, not a customer benchmark, and it says nothing yet about your cameras.
Trust
The platform shows the evidence. People make the call.
An alert, a match and a group are candidates. Each leads back to a still frame where one was captured, an alert stays New until a named person marks it, and the controls that matter are in your hands.
Visit the trust centre- 1Observation · Entrance · candidate raised
- 2Human review · A person decides
- 3Source checked · The still, then your recorder
- 4Mark recorded · Named person, in the audit log
- Purpose
- What each deployment and each sensitive capability is for is agreed with you in writing before a pilot.
- Candidate review
- A match is a candidate until a person compares it with its source.
- Source footage
- A sighting keeps a still where a frame was captured. Video stays on your recorder.
- Audit
- Changes, and searches by plate, scene description and route, are recorded with who made them.
- Retention
- Sighting records and stills are deleted after a period you set. Plate-read history, events, alerts, weekly briefs and uploaded footage have no period yet.
- Permissions
- Five roles, and a list of cameras for each person.
Questions we are asked first
What does VisionObject do?
It reads live camera streams and uploaded footage, detects people, vehicles and carried bags, and records each sighting with its colours, a still frame where one was captured and, where legible, its plate. You search those sightings, review candidate links between them, and work a queue of alerts raised by rules you set.
Which workflow fits us?
There are six: incident investigation, video operations, vehicle search, gate and access correlation, occupancy and queues, and site safety. Investigation on recorded footage is the quickest to start, because it needs no connection to live systems.
What is available today?
Everything this site describes is in the current release, and the capabilities page lists each capability with its prerequisites and known limits. Findings are recorded as notes on alerts rather than in case files, alerts are read in the dashboard and on the event stream rather than sent by email, and face search is not offered. A few capabilities, such as plate watchlist alerts and line and dwell rules, have been verified in automated tests and not yet on live footage in our own testing; their limits say so.
What does it need from us?
A stream address for each camera, or exported clips; a workstation sized with us, or, after assessment, a server or a virtual machine; and named people who will review alerts. Plate reading needs cameras placed so a plate is large enough to read.
How are findings reviewed?
An alert stays New until a named person marks it Acknowledged or False positive, and leads to the still of the sighting that raised it where one was captured. A near-match plate is flagged beside the plate as read. Sightings are grouped only on request, by appearance, and never when they overlap in time. A scene reading is kept apart from the plate reader's own read. Treat a group or a near match as a candidate to compare with its still. VisionObject keeps stills, not live video, so the surrounding footage is checked on your recorder.
How do we evaluate it?
With a pilot on representative footage. We agree one workflow, the measures, such as false-alarm rate and plate read rate, and who reviews. We publish no accuracy figures: the numbers come from your cameras.
Where does our footage go?
Video stays in your deployment. Two optional modules use an AI service that you configure: AI scene reading sends one still of the whole camera view per sighting, with the object type, the plate as read, the camera's identifier, colours, how long the object was followed and the time, and a failed request may go to a second service you configure. The weekly brief sends aggregate figures with the names of your rules, cameras, areas and watchlist labels. Both can be switched off for each organisation. Mail you configure goes through your relay, and the map loads its background from a public map service.
Is face search switched on?
No. Face search is not offered in the current release. It is switched off in every deployment we deliver, and its recognition model is not licensed for production use. An organisation's administrator cannot switch it on; only a change to the deployment's configuration could, and our agreement does not allow it.
Start with one workflow and footage you already have.
Choose the workflow that costs your team the most time. We will walk through it on demonstration footage and plan a pilot on your own cameras, without asking you to send footage through this website.