Video Image Features Extraction Based Smoke Detection Algorithm Research

An important application of smoke detection is for fire warning, because before the fires there are lots of smoke. The traditional fire detection methods are almost based on particle sampling, temperature sampling, relative humidity sampling by ionization or photometry. But the traditional fire detectors are generally used for the situation which has small indoor space and less distance. The detection function is great limited. However, through smoke detection based on video image processing technology can overcome these problems very well, because it is based on technology developed from computer image processing, pattern recognition and machine learning technology. It can detect smoke in a wide detection range and give a rich visual detection information and have fast response to video image information, also have the ability to serve large and open spaces and can provide the abundantly required information as well as can detect fire be away from the fire source.Because of the big advantages about the smoke detection based on video image processing technology in the field of fire detection. This smoke detection methods be-come more and more important. This paper studies the video smoke detection algorith-m based on image features extraction, the research contents include:motion detection based on smoke video and capture smoke region and its features from video images, the algorithm of Random Forest(RF) and Support Vector Machine(SVM) to get a clas-sifier to recognize the smoke block. Based on the previous research, we proposed two different smoke detection methods based on video image process:(1) Based on the diffusion of smoke motion and the semi-transparent of the smoke, we proposed a wavelet energy and smoke growth based video smoke detection method. The strategy of this method include three steps:Firstly, according to the motion of smoke in video, we extract the smoke region in video image by background estimation. Then, since the smoke is semi-transparent, edges and textures of smoke image frames start losing their sharpness and this leads to a decrease in the high frequency content of the image, we respectively divide the background and current image frames into blocks, calculate their high frequency energy in wavelet domain, compare the results, further determine the smoke region. In addition, we analysis the smoke growth rate and perimeter to area ratio of smoke region based on smoke diffusion, and finally complete the smoke detection in video. Experimental results show that the proposed method can detect the smoke quickly and efficiently and also provide an early alarm at a lower complexity.(2) By transforming the problem of smoke detection in video to the problem of smoke pattern recognition in video, and making use of the random forest classification and regression model, we proposed a video smoke detection method with Random For-est features selection. The main strategy of the method is as follow:First, the features we extracted input to RF are color features in RGB space, the wavelet high frequen-cy sub-images, multi-scale local max saturation, multi-scale dark channel. Then using haze image formation model, we make the synthetic smoke images through non-smoke images from offline video, and next we partition these images including synthetic smoke images and the non-smoke images into blocks as the samples for RF. After training the RF with partitioned blocks and their extracted features, we can use SVM to get a clas-sifier which can be recognizing the smoke blocks and the non-smoke blocks. And then the smoke region candidate in video images can be detected by these classifiers. We finally analyze the detected smoke region with the features of the growth rate and the perimeter to area ratio to make the final decision on video smoke detection. The ex-perimental results show that.the proposed method can detect the smoke timely and get a fire alarm in an lower false-alarm rate way; With the RF features selection algorithm can analyze the smoke features automatically; Using synthetic smoke image blocks as the RF samples reduces the complexity of smoke detection algorithm.