Skip to content
VisionObject

Capabilities

Recognition, behaviour, classification and advanced analytics, built on three kinds of detected object: people, vehicles and carried bags. Every capability is listed in full below, with its prerequisites and limits.

Request a demoAvailability register
  1. Detected · Car, followed on this camera
  2. Described · White car, plate read
  3. Seen again · A candidate at the next camera
  4. Kept · A still and a record to search
Illustrative district. Four cameras, four sightings of what may be the same car.

The four capabilities

Four different things

A detection is not a match, and a match is not an identity.

Each page keeps these four apart. A result never moves from one to the next because a score crossed a threshold.

  1. Detection

    An object was observed in a frame or a sequence of frames: a person, a vehicle, a bag.

  2. Appearance similarity

    Two observations resemble one another: by clothing and colour, or by a plate read. Grouping uses appearance only. A plate is searched on its own, and a near-match plate is flagged beside the plate as read. Comparing faces is not offered in the current release.

  3. Reviewed association

    A reviewer accepts that two observations belong together, within one investigation and for its purpose. That decision is recorded as a note on the alert.

  4. Outside the product

    Identity verification

    Establishing who a person is. That is a separate, specifically authorised process with corroborating evidence, carried out by your organisation. VisionObject does not do it.

What is recorded

Attributes that can be observed, and nothing that cannot.

The difference between "a vehicle" and "a white van at the gate at 22:41". Each attribute comes from a model and can be wrong.

Person

  • Clothing colour, upper and lower
  • Descriptive tags, such as "backpack" or "uniform"
  • First and last seen
  • Rule events it raised: zone entry, line crossing with direction, dwell limit reached

No model here classifies demographics or emotion.

Vehicle

  • Type: car, bus, truck, motorcycle, bicycle
  • Colour
  • Plate as read, with confidence
  • Whether the read fits a valid format
  • Rule events it raised: zone entry, line crossing with direction, dwell limit reached

Speed is not measured. A bicycle has no colour and no plate.

Carried item

  • Type: backpack, handbag, suitcase
  • First and last seen
  • Left still beyond the set time, where an item-left rule fired

Limited to these three types. No colour is recorded for a carried item.

Scene reading

  • A one-line description
  • Up to eight tags, from a preferred list that the model may add to
  • Suggested severity
  • The plate as the model read it, kept apart from the plate reader's result

Optional module. A reading can be wrong.

Not inferred

  • Emotion or mood
  • Intent, dangerousness or likelihood of offending
  • Gender, ethnicity or other demographic classification
  • Who a person is, from appearance alone

Availability register

Every capability, with its prerequisites and limits.

One record for each capability. The home page, the platform page, the events catalogue, the industry pages and the release notes all stand on it.

Video sources

What can be connected or uploaded.

  • Live camera streams

    Analyse live streams from IP cameras, recorders and video management systems that publish a standard RTSP stream.

    Prerequisites and limits
    Needs
    A stream address the deployment can reach on your network
    Stream credentials for each source
    Known limits
    Tested on streams re-published by a camera gateway; a direct connection to a camera is confirmed per model and firmware during pilot planning
    The camera is read, not controlled: pan, tilt and zoom and the camera's own events are not used
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Recorded footage

    Upload exported clips one at a time or several at once. They are analysed as fast as the hardware allows and placed on a timeline at the time they were recorded.

    Prerequisites and limits
    Needs
    A start time for each clip: read from the file name where the export includes one, otherwise entered by the operator
    Known limits
    Uploaded footage is never used to raise alerts
    Tested with exported clips from a fixed camera; footage from other sources requires assessment
    A failed analysis job can be re-run; an interrupted transfer is uploaded again
    Playback of some video formats depends on the operator's browser
    The product does not yet show how a clip's start time was established
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Bulk camera registration

    Register a published list of cameras, or a numbered range of stream addresses, in one step and re-sync it later.

    Prerequisites and limits
    Needs
    A camera list in the supported format, or a stream address pattern
    Known limits
    One list format is supported today; others are assessed per site
    Camera locations are read from a list only where it publishes them; the one list tested publishes none, so an operator enters them
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Single-image analysis

    Submit one image through the API and get back the objects, colours and plate found in it.

    Prerequisites and limits
    Needs
    API access
    Known limits
    Only the modules your organisation has switched on are applied
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

Detection

What is found in each analysed frame.

  • Detection and tracking

    People, cars, motorcycles, bicycles, buses, trucks and carried bags are detected in each analysed frame and followed while they stay in view of one camera.

    Prerequisites and limits
    Needs
    A connected camera or uploaded clip
    Known limits
    Detection quality depends on the scene, image quality, light, distance and camera angle
    Only these object types are detected; anything else is not seen as an object
    Not every frame is analysed: frames are sampled, and skipped when the system falls behind
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Still evidence for sightings

    The clearest frame of a sighting, and the full scene it came from, are kept with the camera and the time, wherever a frame was captured.

    Prerequisites and limits
    Needs
    A connected camera or uploaded clip
    Known limits
    VisionObject does not retain live video; the surrounding footage stays on your recorder
    A sighting has no still when its frame was not captured in time, which happens most when analysis is behind
    Stills are not cryptographically sealed
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

Description

What is recorded about each sighting.

  • Colours and descriptive tags

    Optional module

    Dominant clothing and vehicle colours, and up to three short descriptive tags such as "backpack" or "uniform", are attached to a sighting.

    Prerequisites and limits
    Needs
    Appearance module switched on for the organisation
    Known limits
    Colours shift with lighting and camera settings
    Colours are recorded for people and motor vehicles, not for bicycles or carried bags
    A tag describes what a sighting looks like; it is not a fact about a person
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Number-plate reading

    Optional module

    Plate text is read on vehicles that are large enough in the frame, with a confidence value, and settled by a vote across the frames of one sighting.

    Prerequisites and limits
    Needs
    Number-plate module switched on for the organisation
    Cameras placed so that a plate covers enough of the image
    Known limits
    Not every visible plate is readable: small, distant, angled, dirty or poorly lit plates are skipped rather than guessed
    Read validation is tuned to Indian plate formats; other formats are assessed in a pilot
    In our own testing, on night and dawn street footage, only a small share of vehicles produced a usable read; daytime read rates have not been measured
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • AI scene reading

    Optional moduleUses an external AI service

    After a sighting, one still frame is read by an AI vision model, which returns a short description, up to eight tags and a suggested severity.

    One still of the whole camera view per sighting is sent to the AI service you configure, with the object type, the plate as read, the camera's identifier, colours, how long the object was followed and the time. A failed request is retried and may go to a second configured AI service.

    Prerequisites and limits
    Needs
    An AI service configured for your deployment
    AI scene reading switched on for the organisation
    Known limits
    A reading arrives after the sighting; it is not on the real-time alert path
    A reading can be wrong or missing, so it is a candidate for review
    Readings per hour are capped at deployment; beyond the cap, older sightings are skipped rather than read late
    Tags are drawn from a preferred list and the model may add others; rules match tags from the list
    The reading instructions and the tag list are written for city street scenes in India; other settings are assessed in a pilot
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

Association and search

How separate sightings are found and linked as candidates.

  • Grouping sightings across cameras

    Optional module

    On request, sightings that look alike are grouped as candidates for the same person or vehicle across cameras and clips.

    Prerequisites and limits
    Needs
    Appearance module switched on for the organisation
    Known limits
    Groups rest on overall appearance, so similar clothing or similar vehicles can be grouped together
    Sightings that overlap in time are kept apart, so one person seen by two overlapping cameras at once appears as two candidates
    A run covers the 500 most recent sightings in your view, and group labels change between runs
    Tuned on footage from one camera; matching across cameras has not been measured
    A group is a candidate, not an identification
    Accepting, rejecting or splitting a group is not recorded in the product yet
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Vehicle route by plate

    Optional module

    The readable sightings of a plate, in time order, as numbered stops on a map.

    Prerequisites and limits
    Needs
    Number-plate reading
    Camera locations entered by an operator
    Known limits
    Verified by automated tests; the only route recorded on live footage in our own testing had one stop
    Reads that differ by one character or a commonly confused character are included and flagged as near matches
    A stop without a camera location is listed but not drawn on the map
    Lines between stops are inferred: the route between cameras was not observed
    Supported in
    Workstation pilot, On-premises server, Your cloud account

Rules and events

What turns a sighting into an event of interest or an alert.

  • Zone entry and item-left rules

    An alert when a followed person or vehicle enters a zone you draw, or when a bag or a stopped vehicle stays still beyond a set time.

    Prerequisites and limits
    Needs
    A zone drawn on the camera's view, for a zone-entry rule
    Known limits
    An event is recorded only when a rule fires, and each firing is an alert
    An item-left rule covers the camera's whole view, not a zone
    Geometry is drawn on the image, so a moved camera needs its zones redrawn
    Rules run on live cameras only
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Zone exit, line and dwell rules

    Leaving a drawn zone, crossing a drawn line in a chosen direction, or staying in a zone longer than a set time.

    Prerequisites and limits
    Needs
    Zones and lines drawn on each camera's view
    Known limits
    Verified by automated tests; not yet confirmed on a live feed in our own testing
    An event is recorded only when a rule fires, and each firing is an alert: a line rule is not a complete log of crossings
    Geometry is drawn on the image, so a moved camera needs its zones and lines redrawn
    Rules run on live cameras only
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Count rules

    A set number of people or vehicles present at the same time in a camera's view.

    Prerequisites and limits
    Needs
    A threshold chosen for the camera
    Known limits
    A count covers the camera's whole view; it cannot be limited to a drawn zone
    A count of detected objects in view, not a calibrated crowd density
    People hidden behind others are not counted
    While the count stays at or above the threshold, the alert repeats after each repeat limit
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Schedules, repeat limits and exceptions

    A rule can run on a schedule, limit how often it repeats, carry a severity, skip named cameras, and be muted automatically on a camera where reviewers keep marking it false.

    Prerequisites and limits
    Needs
    Rules reviewed with you during set-up
    Known limits
    A repeat limit is a time window for one rule on one camera: a continuing situation alerts again after it, and separate incidents inside it are not raised
    Automatic muting needs alerts to be reviewed; an unreviewed queue mutes nothing
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Rules on AI scene readings

    Optional moduleUses an external AI service

    A rule can fire when a scene reading carries a chosen tag, such as a rider without a helmet or a person on the ground, or a chosen severity.

    One still of the whole camera view per sighting is sent to the AI service you configure, with the object type, the plate as read, the camera's identifier, colours, how long the object was followed and the time. A failed request is retried and may go to a second configured AI service.

    Prerequisites and limits
    Needs
    AI scene reading
    Known limits
    The alert follows the reading, so it is not on the real-time path
    If a reading is skipped or wrong, the rule does not fire or fires wrongly
    The rule editor offers the tags on the preferred list, not tags the model added itself
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Plate watchlist alerts

    Optional module

    A plate on your organisation's watchlist raises an alert when it is read. The alert shows the plate as read, the watched plate and whether the match was exact.

    Prerequisites and limits
    Needs
    Number-plate reading
    A watchlist kept by people with that permission
    Known limits
    Verified by automated tests; not yet confirmed on a live feed in our own testing
    A plate that cannot be read cannot be matched
    A read matches exactly, with commonly confused characters treated as the same, or within one character
    A near match is flagged as such and needs a person to compare the two plates
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Camera health

    Each camera is shown as online, degraded, offline or disabled, and an alert is raised when a feed is lost.

    Prerequisites and limits
    Needs
    A connected live camera
    Known limits
    Health is judged from the frames that arrive, so a frozen but still-streaming picture is not always caught
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Starter rule sets

    A new organisation starts from a rule set for public safety, traffic, campus and facility, venue and retail, or camera health only.

    Prerequisites and limits
    Needs
    AI scene reading, for most of the rules
    Known limits
    Most starter rules fire on an AI scene reading and stay silent without it, including one of the two camera-health rules
    No starter rule uses a zone or a line: those rules are added for each camera
    The rules are written for city street scenes in India; other settings are assessed in a pilot
    Supported in
    Workstation pilot, On-premises server, Your cloud account

Review and response

What an operator can do with an alert or a recording.

  • Alert triage

    Each alert stays New until a named person marks it Acknowledged or False positive. Alerts can be assigned, tagged and annotated; status and assignment can also be set for a selection.

    Prerequisites and limits
    Needs
    A role that may triage alerts
    Known limits
    An alert from a scene reading or a plate watchlist hit carries a still; an alert from a zone, line, dwell or item-left rule links to the sighting and its still; count and camera-offline alerts have none
    The surrounding video is on your recorder
    Tags and notes are added one alert at a time
    Acknowledged and False positive are the only outcomes; there are no incident, escalation or resolution states yet
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Live wall

    One to sixteen cameras with detections drawn on, a plain status for each feed, and a full-screen wall mode.

    Prerequisites and limits
    Needs
    A connected live camera
    Known limits
    Boxes on live video trail the picture slightly; an analysis view shows the exact analysed frame
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Playback of uploaded footage

    Uploaded clips play as one timeline, with the gaps between them shown or skipped, detections drawn on, event markers, speed control and keyboard shortcuts.

    Prerequisites and limits
    Needs
    Uploaded footage
    Known limits
    Covers uploaded footage only; a live sighting has a still, not a recording
    It skips the gaps between clips and never combines objects from different times into one picture
    Skipping gaps shortens playback; it is not a measure of review time saved
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

Reporting

What is counted and summarised.

  • Insights

    Counts by object type and hour, plate-reading volumes, the alert funnel, false-alarm rate by rule and camera health. The headline figures are compared with the previous period.

    Prerequisites and limits
    Needs
    Live cameras or uploaded footage
    Known limits
    A figure for a module that is switched off is shown as unavailable, never as zero
    The false-alarm rate covers reviewed alerts only
    Counts by type and hour cover the last 30 days
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Weekly brief

    Optional moduleUses an external AI service

    A weekly fact sheet for each organisation with a written summary, optionally sent by email.

    Aggregate figures, with the names of your rules, cameras, areas and watchlist labels, are sent to the AI service you configure. Never footage, stills or plate numbers.

    Prerequisites and limits
    Needs
    An AI service and a mail relay configured for your deployment
    Known limits
    The figures are computed first and kept; if the summary cannot be written, the brief says so and shows the figures alone
    Watchlist labels are part of what is sent, so keep plate numbers out of them
    Supported in
    Workstation pilot, On-premises server, Your cloud account

Access and governance

Who can see and change what, and what is kept.

  • Organisations and camera sharing

    Several organisations can use one deployment. Each camera has an owner and can be shared with named organisations; each organisation sees only its own rules, alerts and watchlists.

    Prerequisites and limits
    Needs
    A platform administrator who assigns cameras
    Known limits
    One deployment at a time: organisations on separate deployments are not joined
    A platform administrator sees one organisation at a time
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Roles and camera scopes

    Five roles (administrator, supervisor, operator, investigator and viewer) and, for each person, an optional list of the cameras they may see.

    Prerequisites and limits
    Needs
    Accounts created by invitation
    Known limits
    Roles are fixed; custom roles are not available
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Accounts and sessions

    Sign-in by email and password after an invitation, automatic sign-out after inactivity, and a lock on an address after repeated failed attempts.

    Prerequisites and limits
    Known limits
    Sign-in is within the product; there is no single sign-on through an identity provider
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Audit log

    Changes, and searches by plate, scene description and route, are recorded with who made them and when, and can be exported.

    Prerequisites and limits
    Needs
    A role that may read logs
    Known limits
    Kept for 365 days by default
    Filtering sightings by type, colour or camera is not recorded as a search
    The log records that an action happened; it is not a tamper-evident seal
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Retention settings

    By default sighting records and their stills are deleted after 90 days and audit entries after 365 days; a platform administrator can change both periods. Frame-level detection data is deleted after 30 days, which is fixed.

    Prerequisites and limits
    Needs
    A platform administrator
    Known limits
    Plate-read history, events, alerts and their notes, weekly briefs and uploaded footage have no retention period and no delete operation in the current release
    Sighting records cannot be kept for less than 31 days
    Recordings on your own recorder follow your recorder's settings, not these
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Module switches per organisation

    Number plates, appearance, AI scene reading and the weekly brief are switched on or off for each organisation.

    Prerequisites and limits
    Needs
    A platform administrator
    Known limits
    A switch changes what runs and what can be read; it does not delete what was already stored
    Face search is not among the switches you can use: it is not offered in the current release
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Organisation allowances and licence

    For each organisation a platform administrator can set how many people and enabled cameras it may have, and a licence expiry date. An organisation whose licence has expired is locked until it is renewed.

    Prerequisites and limits
    Needs
    A platform administrator
    Known limits
    Lowering an allowance stops growth and takes nothing away
    While an organisation is locked its rules, watchlists and weekly brief stop, and its own cameras are analysed only if another organisation still sees them; nothing is deleted
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • HTTPS for the dashboard

    The dashboard can be served over HTTPS, set up when the platform is deployed.

    Prerequisites and limits
    Needs
    A host name for the deployment
    For a publicly trusted certificate, a public address for that name that can be reached from the internet
    Known limits
    Verified with a locally issued certificate only; a publicly trusted certificate has not yet been exercised in our own testing
    Uses an automatically issued public certificate or a self-signed one; a certificate you supply is not supported yet
    Off by default; the unencrypted ports stay reachable unless they are closed at the host
    Supported in
    Workstation pilot, On-premises server, Your cloud account

Integration

How other systems read from and write to the platform.

  • API

    Programmatic access to cameras, sightings, events, rules, alerts, plates and reports. A person's calls carry that person's permissions.

    Prerequisites and limits
    Needs
    An account, or the deployment's automation token
    Known limits
    One automation token exists per deployment: it has platform-administrator rights across all organisations and names the organisation it acts in on each request. For role-limited access, use an account
    Every data call after set-up is authenticated; the health report and performance figures are open to anyone who can reach the API port
    One version is published; changes are announced in the release notes
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Real-time event stream

    Detections, events and alerts pushed to a connected client as they happen, limited to what that client is allowed to see.

    Prerequisites and limits
    Needs
    An authenticated connection
    Known limits
    Live only: nothing is replayed, so a client that disconnects reads what it missed from the API
    A slow client loses messages rather than slowing others
    Supported in
    Workstation pilot, On-premises server, Your cloud account

  • Log export

    The audit log and the system log can be downloaded as spreadsheet files.

    Prerequisites and limits
    Needs
    A role that may read logs
    Known limits
    Alerts and their stills cannot be exported in bulk yet
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

  • Health and monitoring

    A health report for each part of the system, and performance figures in a standard format for the monitoring tools you already run.

    Prerequisites and limits
    Known limits
    Both are open to anyone who can reach the API port, which must be kept off untrusted networks
    The figures describe the system, not the content of any footage, and include camera identifiers
    Supported in
    Workstation pilot, On-premises server, Your cloud account, Recorded footage only

Reviewed on 3 October 2026 against the product as it stood on 29 September 2026. Operator screens are described as in the dashboard build of 29 September 2026. The dashboard is being rebuilt; which screens a deployment includes is confirmed in pilot planning. A change is made here first, and every page that mentions the capability follows.

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.

Request a demoRead the pilot guide