Solutions / AI Video Analytics
AI Video Analytics
Nobody Is Watching. The Cameras Are.
Detection runs continuously on the cameras already installed. Events raise an alert in real time, a workflow adds what the scene actually showed, and the footage stays searchable afterwards.
The analytics loop
Trusted where the stakes are high
Pick your estate, see what the cameras already show
Plants, lines and yards. PPE, hazard zones, forklift proximity and line state read from the estate already covering the floor.
- Bought at corporate EHS. One programme across plants rather than a pilot that dies at one site.
- Line state without integration. Running, stopped, blocked or unattended, for machines never wired for telemetry.
- The clip, not the shift. Event-driven recording keeps the incident.
Substations, plant and remote sites. Perimeter, restricted areas and equipment condition where nobody is standing.
- No rip and replace. RTSP and ONVIF over any VMS.
- Fire and smoke on camera. Detected visually rather than waiting on a fixed-point sensor.
- Inside the boundary. Self-hosted inference where footage cannot leave.
Depots, platforms and warehouse floor. Forklift and pedestrian conflict, dock activity, and the incidents that become claims.
- Forklift and pedestrian. Both detected, and the crossings where they meet.
- Custody for the claim. Incident media held with chain of custody.
- Cross-camera. One vehicle or person followed between cameras.
Sites that change weekly. PPE, exclusion zones, equipment movement and condition across a moving estate.
- Per project or portfolio. Stands up on general conditions, standardises across sites.
- Change over time. Condition compared across passes, not judged from one frame.
- Evidence for the dispute. Findings held where a claim can reach them.
Streets, facilities and fleets. Plates, watchlists and zone rules on infrastructure already deployed.
- ALPR and watchlists. Plate recognition with enrolment and alerting.
- Air-gapped where required. Full deployment with no external calls.
- Audited access. Per-camera access control with a record of who viewed what.
Campuses are camera estates too. Occupancy, congestion, space utilization and after-hours activity read from the cameras a district or university already runs.
- Space and flow. Occupancy, hallway congestion and facility usage without tracking individuals.
- Event operations. Crowd density and movement during games, assemblies and gatherings.
- Safety has its own page. Detection-to-response for schools lives on School & Campus Safety.
What You Get
What continuous analytics on existing cameras gives you
Detection that does not blink
People, vehicles, PPE, plates, weapons, fire and smoke, zone entry and wrong-direction movement, read continuously rather than by someone watching screens.
Runs on the estate you have
RTSP and ONVIF over existing cameras alongside whatever VMS or NVR is already in place, with camera downtime surfaced rather than discovered later.
Context after the alert
The detection is immediate and deterministic. A multi-modal model then describes what the scene showed, which reaches situations no fixed class covers.
Anomalies nobody defined
Because the analysis describes rather than classifies, it reaches unusual or ambiguous conditions that were never modelled in advance, including on production and packaging lines.
The archive answers questions
Ask across recorded footage in plain language and get the clip back with the answer, instead of scrubbing timelines.
Events become records
Media around an event is retained with chain of custody and retention policy, so an investigation or a disclosure is answered from the system.
Trained on your own footage
Where a condition recurs and matters, a detection model is trained against your material rather than waiting on a vendor release.
Stays inside your boundary
On-premises, private cloud or fully air-gapped with self-hosted inference, for estates where video does not leave.
How It Works
Detect, understand, respond
Detect
Computer vision reads the live streams and raises an event the moment a rule is met. Real-time and deterministic.
Understand
The event triggers a workflow that reads the cached frames around it and describes what was happening, so the alert carries context rather than a bare class name.
Respond
The workflow decides what follows: notify, escalate, route to a person, or hold the media as a record with custody intact.
Products Inside
The products inside
Nexus
Enterprise content platform
The record underneath: ingest, storage, search, identity, retention and chain of custody.
Explore Nexus →AI Live Insight
Real-time video analytics
Detection on the cameras already installed, over RTSP and ONVIF, with event-driven recording.
Explore AI Live Insight →AI Intelligence Hub
Agentic AI platform
The workflow that runs when an event fires: scene interpretation, cited answers, human checkpoints.
Explore AI Intelligence Hub →FAQ
AI video analytics, asked and answered
What is AI video analytics?
Software that reads camera streams continuously and raises an event when a condition you define occurs -- a person in a restricted zone, missing PPE, a plate on a watchlist, smoke, a vehicle going the wrong way. It replaces someone watching screens, and it runs on the cameras already installed rather than requiring new ones.
Does it need new cameras or a new VMS?
No. Detection runs over RTSP and ONVIF against the existing estate and works alongside whatever VMS or NVR is in place. Camera health and auto-reconnect mean you find out when a camera drops rather than discovering it during an investigation.
What can it detect?
People, vehicles, PPE, faces, licence plates, weapons, fire and smoke, activity, zone entry and wrong-direction movement, with per-camera configuration and cross-camera re-identification. Where a condition matters and is not covered, a detection model can be trained on your own footage.
Is the analysis real-time?
The detection is. Scene-level description runs on cached frames on a cadence, so it tells you what has been happening rather than reading the current frame. The alert does not wait for it.
What happens after an alert?
The event triggers a workflow. It can notify, escalate, route to a person for review, or retain the media as a record with chain of custody. Confidence thresholds decide what is acted on automatically and what a person sees.
Will it fire on something already in progress?
No. A workflow triggers on an event, not on a state that was already true before it was watching. And the response inherits the accuracy of the detection that raised it.
Can it catch things nobody defined in advance?
To a degree, and this is where a describing model beats a class list. Rather than matching against enumerated categories, it describes what a scene showed, which reaches unusual or ambiguous situations -- including on production and packaging lines. It is not a substitute for purpose-built inspection equipment where repeatable measurement against a specification is the requirement.
Can we search recorded footage afterwards?
Yes. The recorded estate is searchable in natural language, and answers return the clip they came from rather than a timestamp to go and find.
Can it run without sending video outside?
Yes. Self-hosted inference supports on-premises, private cloud and fully air-gapped deployment.
Point it at one camera estate for a week
Pick a site and a question. We will run detection over your own cameras and show you what a week surfaces.