Fire & Smoke Model
Model Purpose
The Fire & Smoke model detects real fire and smoke from camera images. It is intended for safety and security monitoring in areas where fire or smoke can be clearly seen by the camera.
This model is not suitable for every situation. Review the sections below to understand its accuracy, expected behavior, and suitable or risky scenarios before using it.
Model Accuracy
| Object | Minimum Size Relative to Frame | Accuracy | Nighttime Performance |
|---|---|---|---|
| Fire | 0.1% | >= 95% | Requires color images at night. Night accuracy is ~70-80% of daytime. |
| Smoke | 1.5% | >= 95% | Hardly detectable at night. |
Detection Behavior
- The model is trained on real fire and smoke data.
- Similar visual sources such as lighters, fire extinguisher smoke, stage effects, clouds, fog, still images, or similar artificial fire/smoke are either excluded or detected with low confidence. This helps reduce false alerts.
- The model analyzes a set of frames to verify whether fire or smoke is real.
- The model also combines detection results with user-configured parameters before triggering an event.
- It typically takes under 30 seconds for the AI to gather sufficient data and generate a reliable event.
Suitable and Risky Scenarios
Use this model when the camera has a clear view of the protected area and fire or smoke can be visually detected with enough size and contrast.
| Condition | Suitable when... | Risky when... |
|---|---|---|
| Camera view | The camera faces the protected area and the view is not blocked. | The view is blocked, unfocused, too far away, or includes many unrelated zones. |
| Image quality | The camera has good resolution, focus, and light sensitivity. | The image is blurry, low resolution, too dark, or affected by glare. |
| Visibility | The area is clear and visibility is good. | Rain, fog, darkness, or poor visibility is present. |
| Object size | Fire or smoke is large enough in the camera frame. | Fire or smoke is too small, too far away, weak, or almost static. |
| Contrast | Fire or smoke stands out from the background. | Fire blends with red/orange backgrounds, or smoke blends with black/white backgrounds. |
| Lighting | Lighting is stable, and nighttime fire detection uses color images. | Strong glare, low light, or nighttime smoke detection is required. |
| Environment | The scene contains real fire or smoke. | Flags, stage effects, artificial smoke, clouds, fog, or similar visuals may cause false detections. |
Low contrast between fire/smoke and the background reduces detection accuracy. In these cases, a larger or closer fire or smoke source may be needed for reliable detection.
Low Contrast: Fire color blends with background
High Contrast: Fire color stands out clearly
Tuning Note
Increase the model score if the task triggers too many false detections from flags, glare, fog, or artificial smoke. Decrease the model score only if real fire or smoke is missed in a suitable camera view.
For general guidance on model score and sensitivity, see Model Accuracy, Score, and Sensitivity.