Flame Detection
Flame detection is an AI recognition algorithm that uses deep learning to detect the presence and location of flames in images or video in real time. Trained on large-scale fire image datasets, it enables early fire discovery and alarm notification for both indoor and outdoor environments.
Algorithm Overview
Camera video is continuously analyzed, with the AI automatically recognizing flame characteristics (color, shape, and motion patterns). It can detect initial-stage fires and outdoor flames early, where conventional smoke detectors and heat detectors tend to lag in detection.
Performance Metrics
| Dataset | Detection Accuracy (mAP@0.5) |
|---|---|
| FIRE | 0.86 |
Edge AI Board (RV1126B) Execution Efficiency
| Algorithm | Processing Time |
|---|---|
| fire_detect | 64ms |
Key Features
- Early detection: Visually detects flames faster than smoke or heat detectors
- Wide-area monitoring: Covers a wide area with a single camera
- Low false detection rate: Reduced false detections from lighting changes, reflections, heating appliances, etc.
- 24-hour operation: Day and night operation when combined with night vision cameras
Use Cases
- Fire prevention monitoring in factories and warehouses
- Early fire detection in forests and outdoor facilities
- Fire detection inside tunnels
- Fire prevention measures in kitchens and restaurants
- Fire monitoring at waste treatment facilities
- Abnormal heat detection in substations and electrical rooms
Edge AI Board Implementation
Using the RV1126B NPU, high-speed flame detection is achieved at 64ms. It can be installed as a camera-integrated edge AI unit and is capable of autonomous detection and alarming even during network disconnection.