Research On Video Based Forest Fire Smoke Detection System

Video based fire smoke detection technology judges fire disasters by analyzing,processing and identifying the smoke images,and this is a non-contact detection technology.It is widely used due to the advantages such as wide range of applications,fast speed and high accuracy,which make up the lacks of traditional detection technologies.How to apply this technology to the complex forest environment has become a research hotspot in recent years.This thesis focuses on the study of video based forest fire smoke detection technology,and the main contributions are listed as follows.Firstly,the algorithm of moving target detection based on background modeling is studied and improved.This thesis integrates LBP operator into traditional codebook model,uses LBP histogram vector instead of original RGB vector and records the local texture information through codebook to improve the adaptability of the illumination change.In addition,in order to improve the running speed of algorithm,this thesis adopts modeling object using block instead of pixel and also adjusts the codeword position.The speed of modified algorithm is significantly improved.Secondly,the characteristics of smoke are analyzed and extracted.Smoke does not only have a single feature but multiple characteristics when it spreads.Through the comparison and analysis of many experiments,5 kinds of characteristics of smoke are extracted including color feature,background blurring feature,irregular contour feature,direction of motion feature and center of gravity moving feature.13 smoke feature values are extracted to form the feature vector which is used for training and recognizing for the classifier.Thirdly,based on random forest algorithm,the smoke classifier is designed.The factors that affect the classification performance including number of decision tree and characteristic variable,data preprocessing method,voting mechanism of algorithm are analyzed.On the basis of experiments,the classifier is optimized,and the classifier with the optimal recognition effect is obtained.Finally,the forest fire smoke detection system is designed by combining the image preprocessing algorithm,moving target detection algorithm,the characteristics of smoke image and the optimal random forest based classifier.The performance test of smoke detection system is carried out through multiple videos and the results show that the system can identify smoke quickly and accurately.