A fire that starts small in an industrial setting rarely stays small for long. By the time a smoke detector trips, you're often looking at a damage bill in the millions and, in the worst cases, lives lost. The detectors weren't wrong. They were just late.
Securade.ai HUB tries to close that gap. It watches your existing CCTV feeds with computer vision models that can pick up the visual signs of a fire (flickering, smoke, weird heat patterns on a thermal feed) often well before a ceiling-mounted sensor sees enough particulate to alarm.
This piece walks through how the detection actually works, where the alerts go, and a couple of real situations where the early warning made the difference between an incident report and a disaster.
Why "see the fire" beats "sense the smoke"
Traditional smoke and heat detectors trip after enough smoke or hot air has accumulated near the ceiling. That's a couple of minutes of headroom you don't have. A camera, on the other hand, can see the source: the first flicker of flame, the wisp of grey smoke curling up from a faulty switchgear cabinet, the temperature anomaly on a thermal feed.
The model behind HUB has been trained on a lot of fire and smoke footage from real industrial environments. That training matters because false positives are the killer of any fire system. Steam from a coffee machine, exhaust from a forklift, sun glare on metal: a poorly trained model alarms on all of them. A well-trained one doesn't.
The detection also gets better over time. Every site has its own visual quirks. The model picks them up as it sees more of your particular floor, your particular lighting, your particular machinery.
What the live video analytics actually look at
In practice, the system is looking at three classes of signal in a video frame: flame (high-saturation orange-red patterns with characteristic flicker), smoke (slow plumes with low-frequency motion patterns), and thermal anomalies on any IR feeds you have plugged in.
It runs across whatever you have: standard IP cameras under fluorescent light in a warehouse, a thermal camera pointed at a switchgear room, a dome covering an outdoor fuel depot. The detection holds up reasonably well under the kind of lighting variation you actually get on a site.
There's also a stack of filters on top to suppress motion that isn't fire. Conveyors moving, vehicles passing, shift changes, those all get filtered out before anything fires an alert. Fewer false alarms, more trust from the people on the receiving end of the pager.
Fusing in other sensors
Video is the centre, but it's not the only signal. HUB can take inputs from regular smoke and heat detectors, gas sensors, and environmental sensors like temperature and humidity. When two or three of these light up at the same time, you have a much higher-confidence signal than any one of them on its own.
This matters most in chemical plants and similar facilities where the failure modes are weird. A gas leak might show up on a gas sensor first, then on a thermal anomaly, then on visible flame, all within a few seconds. Watching only one sensor type misses the early stages.
Sensor configuration is per-site. You pick which inputs are relevant for your environment and tune the thresholds.
What happens when an alert fires
Detection on its own is half the story. The other half is making sure the right person hears about it within seconds, with enough context to act.
When HUB sees something that crosses a threshold, alerts go out over the channels you set up: email, SMS, a webhook into your incident management tool. Each alert carries the camera location, a still frame of the event, a short clip, and the model's confidence score.
If you have it wired in, HUB can also drive your existing fire panel, trigger sprinkler interlocks, or shut down machinery in the affected zone. That's the kind of integration that turns a 60-second response into a 5-second response.
Alerts shaped to severity
Not every detection is the same. A small flame near a welding station is expected; a flame in a paint store is a five-alarm situation. HUB lets you set up tiered alert protocols so the response matches the severity.
A typical setup looks like this. Low-confidence detection in a low-risk zone goes to security for a visual check. High-confidence detection in a high-risk zone triggers the fire panel, kills the local interlocks, and pages the duty manager all at once.
There's also escalation built in. If an alert isn't acknowledged within a configurable window, it escalates up the chain. Nothing falls through the cracks because someone was in a meeting.
Sitting alongside what you already have
HUB doesn't try to replace your fire panel or your sprinkler system. It sits next to them and feeds them better information, faster.
The integrations cover the usual suspects: fire alarm panels, sprinkler controllers, building management systems, access control. The wiring is event-based, so HUB tells your existing systems "this is happening here" and they do what they were already designed to do.
In a real fire event, that translates to: HUB sees the flame on camera, fires the fire panel, the alarm sounds, the sprinklers come on in that zone, and security gets a live link to the camera that saw it. All inside a few seconds.
Two situations where this paid off
A pair of incidents from real deployments, sanitised.
Warehouse, 3am, faulty outlet
A large distribution centre storing palletised consumer goods had HUB watching its cameras. About three weeks after deployment, a faulty wall outlet started a small fire behind a row of pallets at 3am. No one was on that aisle.
The model picked up the smoke before the ceiling sensor did. Alert went to the on-site security team who were on the floor with an extinguisher inside 90 seconds. The damage stopped at one scorched pallet and a melted outlet. Without the early visual catch, that fire had clear runway into the rest of the warehouse.
Data centre, thermal anomaly
A colo data centre had a thermal camera feeding HUB in addition to regular CCTV. The model picked up an unusual heat signature on the back of one server rack: a cooling fan had failed and the rack was slowly cooking.
The duty engineer got an alert, walked to the rack, swapped the fan. No fire, no thermal event, no downtime for the customers in that suite. The fan failure wouldn't have tripped any smoke detector for hours.
Adapting HUB to your site
No two industrial sites look the same. A petrochemical plant has different cameras, different sensors, different rules, and different risk profiles from a tier-3 data centre. HUB is built to be configured per site rather than shipped as a one-size product.
You pick the cameras and sensors you want connected, set the alert routing and severity thresholds, and tune detection rules to your environment. None of that requires writing code; it's all configuration. For sites that already have fire safety protocols in place, HUB plugs into those rather than replacing them.
Camera and sensor options
HUB works with standard IP cameras (most of what you already have), thermal cameras (worth their cost in switchgear rooms and battery storage areas), and special-purpose cameras like high-FPS for flame detection in fuel handling.
On the sensor side, you can wire in heat detectors, smoke detectors, gas sensors, and basic environmental sensors. Bring whatever you've got. The fusion logic gets more confident as more inputs corroborate each other.
Sensitivity, camera resolution, and device placement are all tunable per site. If you also run AI-driven security systems for intrusion or access control, HUB can sit alongside them on the same hardware.
Deployment: edge, cloud, or both
HUB can run entirely on-premise (typical for sites with strict data residency rules), in the cloud (lower friction for smaller deployments), or in a hybrid setup.
Cloud is the easiest to operate because there's no on-site box for you to babysit. On-premise gives you full control over the footage and is what most regulated industries end up choosing.
Hybrid is what most large estates land on: live detection runs at the edge for low latency, aggregate analysis and model retraining happen in the cloud.
Fire detection has been a solved problem in theory for decades. The issue has always been latency: by the time enough heat or smoke reaches a ceiling sensor, you've already lost time you needed. Computer vision shortens that loop by looking at the visible signs first.
HUB isn't magic. It won't replace your sprinklers or your fire alarms. What it does is give your existing safety systems several extra seconds of warning, which on industrial sites is often the difference between a write-off and a non-event.
As sites get more complex and more automated, that earlier warning becomes more valuable. The technology to do it is here. It's mostly a question of plugging it into what you already have.
HUB is open source. The code lives at github.com/securade/hub, and a star helps other people find it. If you want more context on adjacent topics, the resources section has more.