A vision-based system for early fire detection

This paper presents a video-based early fire detection system. Focus is given to the challenges related to the actual deployment of the vision system. Most importantly, background subtraction is performed in a windowed way for improved accuracy, an attentive mechanism is employed to focus a computationally expensive frequency analysis of potential fire regions, interaction with a people detection and tracking system is included so as to enable model-based false alarms rejection, a new colour-based model of fire's appearance as well as a new Wavelet-based model of fire's frequency signature are proposed, and the camera-world mapping is approximated according to a GPS-based learning calibration process. An average success rate of 92.7% at a processing rate of 10 Hz shows the applicability of the model to real-life applications.