The Research On The Fire Smoke Detection Study Based On The Video Sequence Under The Indoor Conditions

With the rapid development of the modern society and the high-rise buildings, the underground buildings, the large buildings, and a variety of new combustible materials being widely used as well as the wealth in society and people highly concentrated, the fire detection technology is devoted to the increased attention because of the fire situation is more and more severe, and especially the fire safety issues in the large space premises are increasingly concerned. Despite the conventional fire detection technology has its own advantages and some technologies are already mature in the long process of development, it has also its own insurmountable disadvantages, especially in the large indoors space. Therefore the fire detection technology based on video sequence gradually begins to develop in the large indoor space due to the image collecting and the rapid development of computer technology.The existing video-based fire detection technology mostly serves for the flame detection and identification. Although it has an initial maturity and some of them have used for practice, its higher cost, smoldering fire alarm problems and the reliability are needed to be improved. In this case, these issues mentioned above can be perfectly solved by proposing fire smoke detection technology and the combination of the flame and fire detection technology for the fire detection.This thesis will devise the computer image processing technology and the image recognition which detect automatically the abnormal fire smoke out and sound the alarm happened in the monitored fixed scene fires through using the Video Surveillance System for the application background. This thesis mainly makes the research on the three features of the extraction of the smoke suspected area, the fire smoke extraction and smoke in the premise of saving time and space and for the purpose of the fast and accurate completion of the fire smoke detection.1,The extraction of the smoke suspected areaThe part of the smoke suspected area pronounced in the video sequences of smoke is needed to be analyzed in order to achieve the purpose of smoke detection and alarm, that is to say, the smoke suspected area is needed to be analyzed before settling it, that is, it is extracted from the image segmentation which measures a variety of the possible targets and uses of the images. This requires a good method for the extraction of the smoke suspected area, which is done a lot of researches and imposed many good ways among which using motion characteristics for extracting the smoke suspected area is more representative. From now on, the following two methods of the moving target detection technique are commonly used:optical flow and background subtraction. The biggest drawback of the optical flow method lies in the complicated calculating method which requires the special hardware support to achieve the real-time requirements. The background subtraction method is frequently applied at the relatively static occasions, which doesn't require high complexity and fast calculation and is able to provide the most complete data features without pre-acquisition information. Therefore, this thesis adopts the background subtraction method in the moving target detection for extracting the smoke suspected area.Background subtraction method is the large branch of the Detection of Moving Objects among which contains a lot of different algorithms which are divided into pre-processing, Background Generation, Foreground Detection and post-processing of the sports areas. The Pre-processing serves for reducing the size of each video image and spatial filtering and so on as well as prepares for the background modeling and the foreground detection and the smoke test; The sports areas prepares for removing a small amount of noise in the foreground image and smoke testing; The background modeling method is the core in the background subtraction algorithm and the different backgrounds modeling methods are correspondent to the different foreground detection methods.Background modeling algorithm includes a lot of options among which the neighboring frame subtraction, running average and mixture Gaussian model are chosen to be researched because their features of the less memory, the simpler algorithm and the real-time requirements. Neighboring frame subtraction forms easily the empty region for the extraction of the smoke suspected areas, which cannot be accurately detected and even no smoke detection. The qualities of the background images and foreground images gotten by the running average and mixture Gaussian model background modeling method in the indoor conditions are the similar for getting the better smoke suspected area. However, mixture Gaussian background modeling is relatively complex, so this thesis chooses the running average as the extraction of the smoke suspected areas.This thesis improves the running average, which is updated in accordance with the running average formula when the current video frame pixels as background pixels and is updated in accordance with the average increment when the current video frame pixels detection as the foreground pixels. The difference image threshold segmentation study is presented, which meets the difference image mean as 0 and the difference as theσnormal distribution and obtains the perfect effect of the smoke suspected areas with the threshold segmentation study.2,The extraction of the fire smoke features1) The color characteristics of the smoke:the particles are relatively small in the early stages of the formation of the fire smoke, where the smoke reflects the light blue and the tiny particles movement is gradually intensified with the temperature gradually increasing and collide with each other, combining into the relatively large particles, where the smoke reflects the light gray. In the most cases, blue and gray can be considered as an important feature for judging the smoke.2) The protruding features of the smoke:the smoke released in the early fire constantly expanses under the effects of temperature and pressure. Due to the internal pressure is bigger than the external pressure, the smoke images shows the protruding features of the flat image. The ration of the area of the smoke suspected areas and the area of the smoke suspected areas with the circumference is considered as a standard to measure the protruding degree of the smoke.3) The characteristics of smoke diffusion:the smoke mixture constantly expands under the higher outside temperature and pressure. This expansion reflects the growing area in the plane graph, which becomes a standard of measuring the smoke.4) The unchangeable characteristics of the smoke point:the development of smoke gradually expands generally started by the ignition source, that is, the smoke point doesn't change, which becomes the standard of measuring the smoke. 5) The characteristics of the main direction of the smoke movement:fire, the smoke generated by fire moves in the all directions under the effects of the temperature difference and pressure difference, where the main direction of the fire movement is up and both sides, forming into the unique shape of the fire, that is, the under part is fine and the up part is rough. This movement is unique, which becomes an important feature of the smoke movement.3,The fusion of the smoke characteristicsEvery single characteristics in the district of smoke mentioned above cannot be used for indentifying the smoke, therefore, these characteristics must be fused together to achieve the purpose for identifying the smoke. However, the relationships among these features are very complicated because they are not necessarily linear and some thresholds are difficult to be determined. In view of these reasons, the traditional methods of the mathematical statistics are difficult to obtain the satisfactory results. Therefore, this thesis adopts the Learning Algorithm of BP Neural Networks for the Feature Fusion to achieve the purpose for identifying the smoke.The extraction of the smoke suspected areas extracts the video image which we need., The extraction of the fire smoke characteristics serves for obtaining the parameters of the fire smoke is extracted from the smoke suspected areas presented in the video images. The fusion of the smoke characteristics makes the integrative judgments for parameters obtained from smoke characteristics to achieve the fire smoke detection.