Research On Methods For Flame And Smoke Detection Based On Video Processing

The use of fire plays a vital role in the human civilization and development process.However,the improper use of fire can lead to a fire event and cause a significant impact on human life and property.In the conventional fire detection technology,the fire detector is easily affected by sensitivity,reliability,stability and other factors,which leads to the fire test results with a lot of limitations.As a new intelligent detection technology,video fire detection technology has real-time,high precision,anti-interference ability and other advantages and is widely used in various fire situations.How to extract key information from the fire video becomes the focus and the difficulties in the fire detection technology research.In this paper,flame and smoke were detected as the main parameters in the fire video,the main contents are:First of all,based on the research of traditional motion detection method,it proposed an improved motion detection method combining adjacent frames and context.The method calculated the threshold of the background point based on the third-order central moments and integrated frame difference method and background difference method to detect the motion for video image sequences.In order to improve the detection accuracy,the use of inter-spatial feature eliminated false motion pixels.Secondly,to further enhance the speed and accuracy of motion detection,we combined with ICA algorithm and FastICA algorithm to study the motion detection algorithm based on GICA.The algorithm was based on the distributed picture of the video images sequence,combining a series of geometric transformation method to achieve the separation of the movement area and the background area.Then,in order to detect fire and smoke within the motion region,taking full account of the characteristics of flame and smoke in the combustion process,we proposed the tests on color,shape,texture and pulsation frequency characteristics of the flame and color,rocking and diffusion characteristics of the smoke.At last,on the basis of the smoke and flames judgment,we researched and analyzed three flame and smoke detection algorithms separately based on simple synthetic,based on the weighted voting mechanism and based on BP neural network.We did experiment in the MATLAB software platform and discovered that the algorithm based on BP neural network has a more integrated consideration of the integration of information from various indicators.Through the continuous training process,the algorithm can achieve more accurate detection results.By studying flame and smoke detection methods in the video enriches the means of video fire detection technology,improves the accuracy of detection algorithm and promotes the development of flame and smoke detection technology in the video.This paper gives the next research direction step.