Recognition And Rearch For Smoking Behavior Based On Video Smoke

Many countries have made anti-smoking laws, smoking is strictly forbidden in publicplaces. Many cities also demand no smoking in public in China, especially the publicplace where a large number of people are there, such as railway station. Because tradition-al smoke detectors are affected by the installation place and the scene environment, theyare also under the influence of the concentration of smoke. They can’t detect the smokeeffectively. In addition, they can’t locate the smokers exactly in the process of smoke de-tection. The above problems can be solved through the smoke detection based video, sowe can effectively take advantage of the monitoring equipment which are used for security,and reduce the burden of monitoring people.This paper mainly discusses smoke criterion about smoking behavior in intelligentvideo surveillance. It can judge the occurrence of smoking behavior by detecting thesmoke. The current environment is indoor environment. Through discussing the sequenceof video images, we can detect smoke and recognize it.This paper looks for the best wayfor smoke detection in the reference of fire. Background subtraction method is used to getthe motion pixel and find the foreground images which are changed. Then the color modelcan distinguish the smoke and other moving objects. The pixel can be found in HSI colormodel, after then they are marked. The smoke suspected area pixel is obtained. Accordingto the clustering feature of smoke, remove the block which is not corresponded to cluster-ing feature.It can avoid the noise which is produced because of the changes of light thatmay affect the detection of smoke. Finally, because smoke pixels in the smoking imagesequence changed over time and the characteristics of translucence, statistical histogram ofpixel and the spatiotemporal characteristics of smoke changes are used to capture the cha-racteristics of smoke. Discuss the luminance component of smoke and color componentsof three channels in RGB model.Make statistics about the captured smoke samples and non-smoke samples, and ex-tract features. The samples are used to train the classifier. The paper chooses support vec-tor machine (SVM) as final classifier. It can recognize the smoke accurately, and the rec- ognition rate is very high.