Research On The Algorithm Of Fire Smoke Detection Based On Video Image

In recent years, with the rapid development of the modern society and the high-risebuildings，the underground buildings, the large buildings, and variety of new combustiblematerials 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 situationis more and more severe. Because of traditional detectors can not overcome someshortcomings, image-based of fire detection technology develops rapidly.The use of video image fire detection, can be overcome the shortcomings of traditionaldetectors, such as space height, air flow may have a greater impact on them, to improve firedetection sensitivity and reliability. Image-based fire detection methods include flamedetection and smoke detection. However, there are no obvious flames when the fire occurredearly, so relative to the flame, smoke detection is better achieve the early detection of firealarm.Based on the summary and analysis of the relevant research works home and abroad, wemake research on how to detect candidate smoke regions and the dynamic characteristics ofsmoke. An algorithm based on digital image processing technology to detect the smoke of thefire is proposed in this paper. The main research contexts and results are as follows:1,This paper improves the running average algorithm，the extraction of the smokesuspected area is based on block. An adaptive threshold based on the image gray scalestatistics is proposed to extract the prospects. A different method is used to update backgroundfor background points and prospect points. Meanwhile, in order to eliminate the impact oflight mutations and other factors, the cumulative number of occurrences of each pixel in theforeground is counted when update the background model. An adaptive threshold is be usedto image segmentation. Combined with the connectivity of the domain area and the colorfeatures of smoke, the connected domains of suspected smoke area are divided to exclude thecoexistence of other moving objects and smoke. Finally, the smoke movement remain isanalyzed combined with the smoke direction of movement.2, Based on the movement direction of smoke, the smoke movement integrity is found after testing and analysis. Smoke often shows the trend of the overall movement in onedirection. Then, the concept of motion deviation is proposed when analyze the connecteddomain of each suspected smoke. Movement deviation refers to the minimum of thedifference between the current block and around the movement blocks. Finally, thecombination of the direction of movement of the smoke, and analysis of smoke movementremain.3, When the environmental conditions are complex, the background is untrue. This isbecause the background at this time often does not reflect the change of scene. It is difficult tobuild an accurate background model influenced by environmental factor in practice. Bycarefully observing background under different conditions, we find that for a simplebackground, non-candidate smoke regions are usually part of the background, their low andhigh frequency energies are equal to background’s. For a complicated background, the energyvalue is large. When smoke does not completely cover background, the average of highfrequency energies in non-candidate smoke regions is largely different from candidate smokeregions. For this reason, we compare the changes of the average energy between smokeregions and non-candidate smoke regions to achieve smoke detection. If the background d issmooth, the change of energy is not obvious. In order to increase the rate of change, weexclude the smooth area in the non-candidate smoke regions. This algorithm is based onwavelet transform analysis.Image pre-process the video images from a fixed camera in the monitoring scene, thenextract the regional campaign with the background subtraction method, and then for eachcharacteristics of smoke, give the method to smoke detection with them, test the segmentationof the target area, if it meets all the characteristics of smoke, then turn to the early warningsystem to determine, obtain the corresponding results.The system is based on the video image. We use the algorithm by detecting the mostrepresentative and most powerful adaptive dynamic characteristics. The algorithm isblock-based, so the complexity of the procedure is low and real-time. The algorithm has awide range of applications, and the common-color smoke can be detected. By using anadaptive background extraction method, the algorithm can reduce the impact of momentaryillumination or lighting changes. When the system find smoke, it warns firstly and alarmsafter five warnings, it could also reduce false positives caused by other confounding factors.Experimental results show that the video based smoke detection algorithm can recognizesmoke effectively and has good anti-interference capability.