CentinelIA360Computer vision for fixed cameras and drones
A single artificial intelligence engine running on the video your cameras already produce, trained for whatever event you need to detect: counting people, monitoring occupancy, spotting vandalism, reading number plates, finding a hotspot or taking a census of swimming pools and solar panels. And it doesn't matter where the image comes from: the fixed cameras you already have on public roads and in your buildings, cameras mounted in a vehicle, an officer's phone, a thermal camera or a drone flight.
A layer of intelligence
on the cameras you already have.
Most organisations don't need another video surveillance system: they need the video already being recorded to be useful. CentinelIA360 sits on top of that infrastructure and turns video into searchable facts — how many people are in the square, which number plate has entered the LEZ, where graffiti has appeared, how many undeclared swimming pools there are in a cadastral area.
One platform, not fourteen products
Ingestion, inference, alerts, evidence and auditing share the same core for every use case. Adding a new detector doesn't require a new deployment or another maintenance contract.
Any camera will do
Fixed on a public road, mounted in a vehicle, an officer's phone, a thermal camera or drone footage. All treated equally, via RTSP and ONVIF. If it produces an image, it works: no civil works and no need to re-tender the capture infrastructure.
Trained for your event
If the detector exists, it is switched on. If it doesn't, it is trained with real images from your environment and validated with metrics measured on your own test set before going into production.
Images stay put unless they need to leave
Inference runs at the edge, on site. What travels outwards is the event metadata, not the video. Face and number plate anonymisation is enabled by default wherever the use case doesn't require identification.
Evidence that stands up in the case file
Every detection leaves a crop, a timestamp, coordinates and an integrity hash, plus a record of who reviewed it and what they decided. Human oversight is documented.
Integrates with what you already use
Output via REST API, webhooks, message queue or GIS export to the video surveillance platform, the map viewer or the municipal case management system.
If it can be seen,
it can be trained.
A sample of the events we detect today, grouped by the unit that usually requests them. None of these lists is closed, nor is it meant to be: what determines feasibility is not the catalogue, but whether the event is visible in the image, whether there is a reasonable angle and resolution, and whether we can gather enough examples to train and validate.
Real-time occupancy against authorised capacity, density map by zone, alerts when the threshold is exceeded, entries and exits at each access point, queues, falls and unusual loitering. And with zone-based pixelation: people are counted without identifying anyone.
Vandalism and graffiti, covert vehicle-based surveillance, illegal dumping, overflowing bins, perimeter intrusion, abandoned vehicles and objects, and the condition of street furniture.
Number plate recognition, LEZ access control, parking space and ORA occupancy, double parking, bus and cycle lane infringements, disabled parking bays used improperly, and vehicle counting and classification.
Swimming pool detection for cadastral cross-checking, solar panel census, unlicensed building work and extensions, roofs in poor condition and uncontrolled tips. From drone flights or orthophotos.
Tree stock inventory, tree health and pest detection, water stress, biomass and fire risk assessment, with comparisons over time between flights.
On thermal cameras: hotspots and incipient fires before there is a flame, embers after a fire has been put out, searches for missing people at night or in undergrowth, and thermography of electrical panels and solar farms.
From image to case file,
in four steps.
The same process for counting people in a square as for finding swimming pools in an orthophoto. The detector changes, not the platform.
Existing IP camera via RTSP or ONVIF, patrol vehicle camera, drone video during or after the flight, orthophotos and previously recorded files. Without replacing the capture infrastructure.
Choose a detector from the catalogue or define a new one with the unit that will use it: what counts as an event, what doesn't, what confidence threshold and which area of the image it applies to.
Labelling on real images from your environment, training and validation against your own test set. The metrics report is delivered before the detector is signed off.
The event is sent via REST API, webhook or to your video surveillance platform, GIS or municipal case file, with its associated evidence and audit log.
It doesn't stop at the alert:
it acts and investigates.
An alert that nobody sees in time is useless. Where the use case allows, the detection triggers the mechanism directly; and when you need to reconstruct what happened, the rule cross-references several cameras at once instead of forcing you to review recordings by hand.
Dry contact and relay
GPIO output or relay module for any device that closes with a contact. It is the common denominator of almost every existing installation.
Turnstiles, barriers and bollards
Opening when capacity is available or for whitelisted number plates, automatic closing when authorised capacity is reached, and bollards that drop for residents and rise again behind them.
Signage and public address
Variable message signs, access traffic lights and public address announcements when an area nears its occupancy limit. Or the command goes to the access control system that already runs the site.
Retrospective search
Across existing footage, without watching it all again: “every white vehicle that passed these four cameras between 21:00 and 22:00”.
Rules across cameras
“If it comes in through the north entrance and doesn't leave through any exit within two hours, alert me.” The condition is evaluated across several cameras at once and reconstructs the route on the map.
With a person in the loop
Opening a barrier can be automated; issuing a penalty cannot. Anything affecting people's rights first goes through an officer who validates it, and every retrospective search query leaves a record of who, what and when.
Seven examples of what it already does.
They are not the limit.
These seven products are proof that the engine works on the street, not a laboratory promise. But they are examples, not the catalogue: the engine is the same for any other event and any other camera. If your case fits one of them, start there; if it fits none, that is exactly what the platform is for. And whatever is detected then goes where the work gets done: the Appolo case file, the Centinel360 resident report, the ProteCivil360 deployment or the FlotaCore360 vehicle.
OCR-ID · Number plates
ANPR reading mounted on patrol cars, motorbikes, drones and portable cases. Stolen vehicles, expired MOT (ITV), uninsured vehicles and LEZ control, with DGT integration. View product sheet.
Grafiti-ID · Vandalism
Deep learning that identifies the author of graffiti from the characteristics of the stroke, with cross-referencing between municipalities and handwriting expert analysis for the case file. View product sheet.
Drone-ID 360 · UAS fleet
Pilot identification and an auditable flight log (SORA 2.5, RD 517/2024), plus a catalogue of AI modules for aerial video. View product sheet.
VIGÍA360 · Covert
Four hidden cameras in an ordinary vehicle, with on-board AI that detects and classifies people and vehicles without an operator watching the screen. View product sheet.
Centinel COP360 · Bodycam
Body-worn and vehicle camera with live 4G streaming, GPS and pre-recording. Video analytics are built into the camera itself. View product sheet.
Aparca360 · Parking
Pay-and-display zones and access authorisations with a police app: on-the-spot number plate OCR and connection to the LPR cameras already installed. View product sheet.
GeoCore360 · Geospatial
The mapping core where detections land: IGN, Cadastre and PNOA, geofences and AI detection across the territory, exposed as a REST API. View product sheet.
Applying AI on public roads
must be justified.
Every deployment is documented before it is switched on: purpose, legal basis, risk category under Regulation (EU) 2024/1689 on Artificial Intelligence, minimisation measures, retention period and who oversees decisions. Uses prohibited by the Regulation are not implemented.
Current certifications · audited by accredited bodies
Tell us what you need to detect.
An adviser will go through the event you want to tackle and the cameras you have, and tell you whether the case is feasible as proposed. If it isn't, we will say so.