Smoke Detection Based On Image And Video

The traditional fire detection have many deficiencies, in recent years, lots of researchers at home and abroad start to find a new method to detect fire, they use digital imageprocessing and video processing to detect fire and smoke in the early of fire. This method's advantage is that it can monitor greater range and find out where the smoke is, Therefore, fire disaster can be controled timely.Smoke is an important feature of flame, it always began to spread before the birth of the flame. Compared with the flame, smoke spread in the whole space, less susceptible to the shelter from object. Smoke detection can help us to detect fire in the early of fire. At the same time, it can also be used to monitor chemical leaks. However, the color characteristics of the smoke are not obvious, and it's motion features are hard to detect. Compared with the detection of flame, it's false detection rate is higher, and it is more difficult to achieve.In order to improve the accuracy of the image based on smoke detection algorithms, and eliminate the similar object of smoke region, reduce the false detection rate caused by environment light changes, a new smoke detection algorithm is presented in this paper. And this paper researches on the following three aspects:(1) Smoke detection based on Video. Using background modeling to filter the background and find out the motion Region in the video, then analyze the color characteristics of the smoke, find out the moving smoke in the video.(2) Extract colormoment feature of smoke. One method the idea of colormoment is that any color distribution in the image can be expressed with its moments, and the information of color distribution concentrated in the lower moments, extract and analyse smoke characteristics in the H,S,V chroma color space, to prepare for future classification. Another method is using the video block accumulation movement and the main direction of motion to detect the smoke.(3) Extract wavelet feature of smoke. Using wavelet transform on smoke image to extract high and low frequency characteristics in different directions, prepare for future classification.(4) Smoke detection based on single image classification algorithm. Using LDA (linear discriminant analysis), SVM(support vector machines), KNN(k-Nearest Neighbor algorithm) to learn these characters, then mark the smoke region in the new testing image automatically. Compare experimental results, and find out the best classification algorithm.