Research On Detection Algorithm Of Video Smoke

Smoke detection based on video for determining monitor whether there is a fire in the scene has the vital significance.Compared with the traditional smoke detection method, video smoke detection has fast response speed, little affected by the environment and direct detection results.In the early stages of the fire will often produce a large amount of smoke, the main research work of this paper is through the video monitoring to detect whether there is a scene in the generation of smoke.This article first to the commonly used three kinds of moving targets detection method has carried on the detailed research.Through experiment contrast the disadvantages and advantages of the three kinds of moving target detection algorithms, and then after the movement characteristics of smoke shown in video analysis, determined the suitable for smoke detection algorithm based on gaussian mixture background model of moving target detection and extraction method.Foreground area is the next step to detect the movement further processing to determine the suspected smoke area.First in RGB color space to color image analysis, set color characteristic criterion ruled out the smoke movement area.Then on after moving targets detection and analysis of characteristics of color area for binarization processing and morphology processing, eliminate the interference of noise in the image point suspected smoke area in video image sequence is obtained.After get suspected smoke area, analysis and extraction of the three kinds of dynamic characteristics of the smoke.Dynamic characteristics including smoke image irregular shape characteristics, smoke smoke image area on the growth characteristics and smoke in the main direction of fuzzy feature in the background.The first to use the actual smoke image on the dynamic characteristics of smoke are analyzed in detail, and the dynamic characteristics of smoke every decision rules, extracting method is introduced in detail, finally verified through the experiment of concrete.Then designs the BP neural network to extract smoke dynamic features of fusion.The design of BP neural network for video smoke detection principle, parameter selection made detailed instructions and experimental validation.For neural network learning and recognition processes are analyzed in detail.Finally, the effectiveness of the algorithm are verified.The first to use this algorithm in different scenarios of multiple video was tested with the smoke, smoke and the testing results are analyzed in detail.Then in the same monitoring scenarios, test results with other smoke detection algorithm by ROC curve are compared.Results show that the algorithm can effectively detect smoke in video image sequence, and has a good anti-jamming.