Fast fire flame detection in surveillance video using logistic regression and temporal smoothing

• Logistic regression is applied to classify the fire and the non-fire regions.
• It significantly reduces the average processing time per frame to 55 msec.
• Temporal smoothing reduces false alarms, with the mean detection delay of 2 sec.

Real-time detection of fire flame in video scenes from a surveillance camera offers early warning to ensure prompt reaction to devastating fire hazards. Many existing fire detection methods based on computer vision technology have achieved high detection rates, but often with unacceptably high false-alarm rates. This paper presents a reliable visual analysis technique for fast fire flame detection in surveillance video using logistic regression and temporal smoothing. A candidate fire region is determined according to the color component ratio and motion cue of fire flame obtained by background subtraction. The candidate fire region is examined for genuine fire flame in terms of the proposed fire probability computed using logistic regression of prominent features of size, motion, and color information. Temporal smoothing is employed to reduce false alarm rates at a slight decrease in sensitivity. Experiments conducted on various benchmarking databases demonstrate that the proposed scheme successfully distinguishes fire flame from the background as well as moving fire-like objects in real-world indoor and outdoor video surveillance settings. The average time to fire detection was fastest among the state-of-the-art video-based fire flame detection techniques for comparison.