Study Of Cigarette Smoke Detection Algorithm Based On Video Surveillance In Indoor Space

Smoking not only causes harm to people’s health, but also makes serious effect on health of surrounding passive smokers. If smoking behaviors can be located accurately when appearing anywhere, we will be able to d etect and stop the smoking behaviors. Compared with manual supervision methods, the means via intelligent video surveillance have a wider monitoring range and the advantage of automatic positioning and alarm function, etc. Besides, intelligent video survei llance can greatly improve the management efficiency in smoking places, because it can intelligently analysis video images of monitored region. In this paper, the characteristics of indoor public places and cigarette smoke are taken account to research cig arette smoke detection algorithm based on video surveillance.Research objects in this paper are the real-time video signal collected by surveillance cameras. First, this paper compares the practical effect of two kinds of background modeling algorithms, then decides to employ mixture Gaussian model background to update background real-timely of the monitored area and combine background subtraction to extract foreground objects in video frames. Secondly, the foreground object images are conducted median filtering and obtained de-noised binary foreground image via applying Otsu to calculate binary threshold value combining with morphological filtering. Thirdly the interesting regions are obtained passing counting projection histogram of binary foreground image in the X, Y direction. Finally, taking account of the detection efficiency of cigarette smoke, the geometric characteristics of the extracted interesting regions are implemented preliminary judgment and the following conclusions are summed up: If the geo metric characteristics suggest it is the cigarette smoke, Hog features of the region are extracted further and cigarette smoke recognition is carried out using the classifier soon afterwards; If the geometric characteristics is not considered as cigarette smoke, the next frame run directly.In order to generate cigarette smoke recognition classifier, this paper uses support vector machine(SVM) with certain robustness as small sample learning machine to train sample set. Sample images were collected through manual control, then above samples containing cigarette smoke and the ones free from smoke cigarette were saved as positive and negative sample set, next extract Hog feature of positive and negative sample set, last the extracted characteristic values were combined into feature vectors to perform decisions classification and generate recognition classifier of cigarette smoke.This paper builds a test platform to test cigarette smoke videos, and verify the timeliness and accuracy of cigarette smoke detection algorithm proposed in this paper. Furthermore, this paper put forward the scope of application and improvement by analysing experimental data.