AI · computer vision · 2025
AICO Vision Fire Detection System
AI that turns existing security cameras into an early warning system that sees flames and smoke, day and night.
We added AI to a building’s existing cameras. Trained on more than 350,000 images, the model recognises flames and smoke in colour footage by day and infrared footage at night, and tells them apart from misleading images like sun glare and headlights. It was tested in a real building with 26 cameras.
Live simulation
Day or night, 6 cameras on one screen.
Watch 6 cameras of a facility live. Start a flame or smoke on a camera and see how many seconds AICO Vision needs to catch it. Try misleading images like sun glare, welding light or headlights; switch to night mode and see it work on infrared video too.
Scenes and detection are simulated for this page; in the real system your cameras’ live video is analysed by an AI model trained on more than 350,000 images. In night mode cameras switch to infrared (IR): flame appears very bright and smoke as a grey haze. An alarm is raised when the probability stays above 60% for a set time.
Challenge
Smoke detectors react too late in large, high-ceiling spaces, and nobody can watch existing cameras 24/7.
Our solution
An AI layer that analyses existing camera footage in real time, recognises flames and smoke day and night, and sends notifications with snapshots.
The story
How we built it
In large, high-ceiling spaces such as warehouses, production areas and car parks, smoke can take minutes to reach the ceiling detector — and in open areas it never does. Yet most of these areas are already monitored by cameras. Our goal was to turn those cameras into fire detectors without any new hardware investment.
We trained the AI model on more than 350,000 images: flames of different sizes, thin and dense smoke, day and night (infrared) footage, indoor and outdoor spaces. We deliberately added examples that look like fire but are not — sun glare, headlights, welding light, orange workwear, steam — and that was the key to reducing false alarms.
The model was tested and refined on the 26-camera system in the building where AICO is based, using controlled flame and smoke scenarios, small flame sources such as candles, and a range of indoor conditions. Because analysis runs locally, detection continues even if the internet goes down; events are recorded and sent once the connection returns.
Technologies used
- Deep learning
- Image processing
- IR imaging
- Local (edge) analysis
- IP cameras
- Live monitoring interface
Technical specifications
| Detection | Flames and smoke, simultaneously |
|---|---|
| Training data | 350,000+ images |
| Footage | Colour by day · infrared (IR) at night |
| Cameras | Existing, compatible IP cameras |
| Field testing | 26 cameras |
| Connection loss | Analysis continues locally, events are stored |
| Notification | With snapshots, via the channel chosen per project |
| Interface | Live monitoring and probabilities per camera |
AICO Vision
Detects flames and smoke day and night using the cameras you already have.
View solutionTry it free with your cameras
If you have a similar need, let’s bring our experience to your project.