A vision-based fire detection system that identifies flames and smoke patterns in real time — reacting faster than conventional sensors and eliminating false alarms.
The challenge
Traditional smoke detectors react after combustion is established. In industrial settings, seconds matter. Existing camera-based solutions are too slow and too prone to false positives.
The approach
Trained a custom YOLO model on thousands of fire and smoke images across diverse environments. Deployed on edge GPUs at camera level for sub-second inference, with a central dashboard for monitoring and automated suppression triggers.
<1s
Detection to alert
99.2%
True positive rate
-96%
False alarm reduction