Computer Vision-based Early Fire Detection Using Enhanced Chromatic Segmentation and
Optical Flow Analysis Technique
Arnisha
Khondaker1, Arman Khandaker1, and Jia Uddin2
1Department of Computer Science and Engineering,
BRAC University, Bangladesh
2Department of Technology Studies, Woosong University, South Korea
Abstract: Recent advances in video processing technologies have
led to a wave of research on computer vision-based fire detection systems. This
paper presents a multi-level framework for fire detection that analyses
patterns in chromatic information, shape transmutation, and optical flow
estimation of fire. First, the decision function of fire pixels based on
chromatic information uses majority voting among state-of-the-art fire color
detection rules to extract the regions of interest. The extracted pixels are
then verified for authenticity by examining the dynamics of shape. Finally, a
measure of turbulence is assessed by an enhanced optical flow analysis
algorithm to confirm the presence of fire. To evaluate the performance of the
proposed model, we utilize videos from the Mivia and Zenodo datasets, which
have a diverse set of scenarios including indoor, outdoor, and forest fires,
along with videos containing no fire. The proposed model exhibits an average
accuracy of 97.2% for our tested dataset. In addition, the experimental results
demonstrate that the proposed model significantly reduces the rate of false
alarms compared to the other existing models.
Keywords: Fire detection, color segmentation, shape
analysis, optical flow analysis, Lucas-Kanade tracker, neural network.
Received
September 26, 2019; accepted March 17, 2020