Research On Indoor Cigarette Smoke Detection Algorithm Based On Feature Fusion

Cigarette smoke contains a variety of harmful substances.Smoking in public places not only causes damage to smokers' health,but also causes a health threat to passive smokers who are around smokers.Regulations on smoking control in public places have long been put on the agenda.Intelligent monitoring system can be used in the indoor and public places because it has the characteristics of quickness and convenience,timely processing and so on.The recognition of indoor cigarette smoke based on video is an effective way to control smoking in public places,which can reduce artificial cost and increase the efficiency of smoking ban.What is more,it has important and far-reaching significance for the future intelligent city construction.In this paper,a video library of cigarette smoke was collected and built through camera equipment to detect cigarette smoke in the video.As to get the foreground region of the smoke in the video,the paper proposes an improved ViBe algorithm and background subtraction method which are combined.Experiments and comparisons show that the algorithm not only improves the processing speed of each frame video sequence,but also eliminates the wrong background area to a certain extent,so as to quickly and effectively extract the suspected smoke area.After the detailed analysis and research on the characteristics of cigarette smoke for the selection of features,it is found that the HOG and LBP features can describe the outline and texture characteristics of cigarette smoke,and these features have a certain degree of anti-interference and the photoperiod invariance.Considering the single feature can not adequately describe the cigarette smoke's characters of small,thin and translucent features,feature fusion method is proposed in this paper.Two kinds of features are fused to form feature vector combined to describe the smoke,in order to improve the reliability of the feature vector,so as to make the expression of smoke characteristics more full and rich.Considering the real-time and stability of the experiment,this paper adopted a classifier designed by the support vector machine(SVM),which is fast,robust andeffective in classification,to classify the smoke.The experimental results show that the algorithm has good accuracy,real time and effectiveness in indoor cigarette smoke recognition.