Fire Detection Based On Video For Indoor Environment

Designing an accurate and real-time fire detection system with indoor localization is of significance in fire prevention.In recent years,various video-based detectors have attracted more and more attention due to their advantages of low cost,high accuracy,fast response speed,and non-contact.So video-based detectors have become the mainstream in the field of fire detection.This thesis studies two working steps of the video-based fire detector: moving region segmentation and target classification.The section of moving region segmentation proposes an improved background subtraction for the complex and variable indoor localization to complete accurate and stable detection of moving targets.The section of target classification uses two methods to identify fire targets.The first is to use artificially designed features and principles to complete fire detection.This thesis selects five features — RGB color features,HIS color features,the first color moment,circularity and histogram similarity — to analyse fire areas.The other method is based on deep learning.Identifying the input picture,this method automatically extracts features distinguished between fire areas and non-fire areas by training a multi-layer neural network and classifies the differences by using Dense layer in the end of the neural network.In this thesis experiments for the above-mentioned algorithms are all designed and conducted.The results prove that the improved background subtraction proposed in this thesis has better robustness and adaptability than other traditional algorithms in a complex indoor localization.This thesis respectively builds complete fire detection systems based on artificially designed features and deep learning methods.Through experiments,the fire detection system based on deep learning is superior in the accuracy and real-time performance.But the advantage of the fire detection system based on artificially designed features is cheap because of the low requirements of hardwares.Therefore,it is necessary to select a suitable fire detection system with consideration of both specific occasions and conditions.