Research Of Smoke Identification Method Based On Video Content Analysis

Fire is a serious threat to the safety of human life and properties.Early detection and early warning of fire are of great significance to reduce various losses.With the popularization of intelligent monitoring equipments,fire detection technology based on video has attracted more and more attention.In the early stage of fire,the object is in the state of smoldering,which is often accompanied with smoke.Conventional fixed-point detectors detect certain particles of smoke by optical or ionization methods.The smoke quantity is required by the fixed-point detectors,and there is a certain time delay when the smoke arrives at the detectors.It is not possible to install fixed point detectors to detect fire due to the random space range of smoke diffusion outside.To some extent,smoke identification with cameras installed in various scenes can solve the problems of outdoor scene application and real-time performance of fixed-point detectors.According to the existing video smoke detection,the moving target detection method is used to segement the foreground region.But there are many interferences and smoke characteristics are not obvious in the foreground region.This paper did the following work:(1)A region segmentation method to extract suspected smoke area was designed.OTSU algorithm was used to segment the image twice to obtain the region of interest containing suspected smoke.Visual background extractor method was used to detect the moving targets in the region of interest and search strategy was carried out on the common area to obtain the suspected smoke area.Experimental results showed that this method could extract more complete suspected smoke area compared with other methods.(2)A SVM smoke recognition method based on multi-feature fusion was designed.For the interference of pedestrians,cars and other moving objects in the scene,smoke motion features such as irregular contour and diffusion were extracted from the suspected smoke area.Extracted smoke static features such as color,blurring background,texture,etc.The model of smoke recognition was obtained by fusing the extracted static and motion features with support vector machine.The method was tested and verified by videos including smoke and non-smoke,and compared with other algorithms.(3)To solve the problem of few public data sets,the smoke video data sets used in this paper were collected on the spot.The region segmentation method designed in this paper was used to extract the suspected smoke area in video,the average precision was 81%.The smoke characteristics in this area were analyzed to train the SVM model based on multi-feature fusion.The experimental results showed that the average accuracy of this method was 91.76%.