CYRA Lab · Module 06
Flame Detection
A flame flickers; an orange object stays still.
Show the camera the flame of a lighter, a candle or a match. CYRA tells a flame apart by its colour and its flicker: being orange is not enough, the flame’s shape must change from frame to frame. When a flame is detected a scenario plays out: the pergola raises an alarm and opens its louvres to let heat and smoke out.
- Runs on your device
- No model download
- Model: Colour and flicker analysis (rule-based) · no model
How to try it
Turn on the camera. Light a candle, lighter or match somewhere safe; do not try it near anything flammable.
Hold the flame 20–40 cm from the camera: the flame-coloured area is marked in amber, and confidence and flicker rise in the panel.
When a flame is detected the scenario starts: an alarm is raised and the virtual pergola’s louvres open to vent. For comparison, show a still orange object: its colour is picked up, but it does not count as a flame.
Your camera permission is used only on this page and only when you turn it on.
Your image never leaves your device: no frame is recorded or sent.
Flame readout
Analysis starts when the camera is on.
Flame confidence
0%
Flicker
0%
Alarms
0
Scenario: pergola
Louvres closed · normal position
Rule-based demonstration: not a fire detector and no replacement for smoke detectors.
CYRA log
- Once you turn on the camera, we start talking here.
Honest note: this module is not a certified fire detector or fire alarm and does not replace smoke detectors; it is a demonstration based on colour and flicker rules. Take care when trying it with a flame; do not try it near children or flammable materials.
Behind the scenes
- About 15 times a second the image is reduced to 128 pixels wide. The flame-colour rule: the red channel dominates in order (red ≥ green > blue); bright, saturated, an orange-to-yellow hue, and clearly brighter than the scene average.
- Colour is not enough: the shape of the flame-coloured area is compared across two frames with their centres of mass aligned (an object moved by hand does not count as flicker), and the change in its area over the last second is measured. Confidence combines the coloured area with the flicker.
- This is not an AI model but a rule-based image analysis. Sunlight, flickering orange lights, screens and moving orange surfaces can mislead it; a small or distant flame can be missed.
Where it helps
- Pergola and awning scenario: a control idea that raises a warning and opens the louvres or retracts the awning when a flame is seen.
- Barbecue and cooking areas: a reminder that an open flame has been left unattended.
- Workshops and storage: an idea for an early, image-based warning layer alongside smoke detectors.
Example use ideas; we design the right one for your project together.
Next module
Silent Talk
Understand each other without a word.
What would you like your project to sense?
Motion, objects, people or surroundings: tell us what you need and let's design the embedded solution together.
Models: Google MediaPipe Gesture Recognizer, Face Landmarker and EfficientDet-Lite0 (COCO dataset), Apache-2.0 licensed. Spoken replies use your browser's own voice.