The Research Of Smoke Detection Algorithm On Video

Fire is one of the major disasters which are serious threats to public life and property safety.The traditional fire detection technology has many deficiencies,such as narrow scope of application,high price,high rate of false positives,slow response and less information of the disaster,etc.Aiming at the existing deficiencies,the fire and smoke detection technology based on video comes into being.The smoke can be detected earlier than the flame based on video,so the smoke detection technology can make the firefighters adopt corresponding solutions and control the fire effectively before the fire spread.Therefore,the research of the smoke detection algorithm on video has important significance and application value.In this paper,we make a detailed research work about the smoke detection algorithm on video.On this basis,we propose a smoke detection algorithm based on the static feature and dynamic feature of smoke.The experimental results show the effectiveness and the robustness of the smoke detection algorithm proposed in this article.The algorithm proposed in this article can be divided into two parts,including the smoke candidate region extraction algorithm and the smoke recognition algorithm.The main research work in this paper is as follows:(1)Research the background subtraction algorithm based on the mean adaptive background modeling.(2)Research the movement region extraction algorithm based on the illumination invariant color representation and photometric gain.(3)Select the smoke candidate regions from the movement regions with the distribution range of the smoke pixels in chrominance(C),intensity(I)and saturation(S).(4)Combining with the theories of the hue-saturation descriptor(HS-Descriptor)and the histogram of oriented gradient(HOG),we propose the smoke detection algorithm based on the hue-saturation background subtraction descriptor(HS-BS-Descriptor).(5)Combining with the theory of the histogram of oriented optical flow(HOF),we propose the smoke detection algorithm based on HOF and HS-BS-Descriptor.(6)Research the theories of support vector machine(SVM)and the evaluation indexes of the classifiers.Train the smoke classifier based on SVM in this article.