Intrusion & perimeter
Line crossing and zone entry with schedules, so a yard that is legitimate at noon is an alert at midnight.
AI-Enabled Video Analytics
Alerts that fire before it becomes an incident
Intrusion, loitering, crowd density, PPE compliance and abandoned objects — detected on the live stream and pushed to the people who can act.
Home › AI-Enabled Video Analytics
Overview
Cameras nobody watches are cameras that record the past
A site with two hundred cameras and four operators is not being watched. It is being recorded, and reviewed after something goes wrong. Analytics changes the order — the system watches every frame, and a person is only pulled in when something needs a decision.
The hard part is not detection. It is keeping false alerts low enough that operators still trust the alarm on day ninety. We tune each rule against your own footage, and we would rather ship four reliable detections than twelve noisy ones.
- Rules tuned on your footage, not a public dataset
- Alerts routed to the operator, app or system that can act
- Every alert stored with the clip that triggered it
- Runs on edge hardware where bandwidth is tight
Capabilities
What it does
Every item below is in the product today, not on a roadmap slide.
Crowd & queue analytics
Density, dwell and queue length, with thresholds that warn before a space becomes unsafe.
Safety & PPE monitoring
Helmet, vest and restricted-zone compliance checked continuously instead of by spot audit.
Abandoned & removed object
Objects left behind or taken away in sensitive areas, flagged with the clip either side of the event.
People counting & flow
In and out counts per entrance, trended by hour so staffing decisions have numbers behind them.
Camera tamper & health
Scene change, defocus, obstruction and stream loss raised as events, not discovered later.
How we deploy
From survey to steady state
Four stages, and we will tell you at the end of any of them if the rest is not worth doing.
Define
We agree exactly what event you want raised, and what the operator is expected to do when it fires.
Trial
Rules run against your recorded footage first, so you see the alert rate before it reaches anyone.
Tune
Zones, schedules and thresholds adjusted until the false alert rate is one your team will live with.
Roll out
Deployed camera by camera, with the alert rate reviewed after the first two weeks.
Specifications
The practical detail
- DetectionsIntrusion, loitering, crowd density, queue, people count, PPE, abandoned object, tamper
- DeploymentEdge GPU box per site or central GPU server — chosen by bandwidth, not preference
- LatencyLive-stream detection. Per-frame latency is measured on your hardware during the pilot.
- Alert deliveryIn-app event feed, email, and push to your messaging or PSIM platform
- PrivacyAnonymised counting by default; identity features are opt-in and logged
- EvidenceEach alert stored with clip, camera, zone and timestamp for review
Questions
Frequently asked
How many false alerts should we expect?
Some, at first. Any vendor claiming zero is describing a demo. What matters is that we trial rules on your own footage before go-live and keep tuning until the rate is one your operators accept.
Do we need new cameras?
Often not. Analytics needs adequate resolution and a sensible angle, which many existing cameras already have. We check during the survey and only recommend replacement where the current view genuinely cannot support the detection.
Does the video leave our network?
Only if you want it to. Edge deployment keeps frames on site and sends events, not video.
Can you add a detection we need that is not listed?
Usually. Tell us the scene and what should trigger the alert, and we will tell you honestly whether it is a tuning job, a training job or not worth attempting.
Read more
More on video analytics
22 video analytics use casesCCTV surveillance softwareWhat is a VMS?
Want to see it on your own cameras?
Send us a camera list and what you are trying to catch. We will tell you what is realistic before anyone talks price.