Wavelet Texture Features Based Video Smoke Detection

Most of the traditional fire detections are temperature sensing type、smoke sensingtype or those two types complex, they have high accuracy for specific points, but it isdifficult to detect fire for large space, large area scene accurately. In recent years, with therapid development of computer technology, image processing and pattern recognition isdeveloped consequently. By using video image processing technology to detect fire, thelimitation of traditional fire detection is solved effectively. Smoke is produced at the earlystages of the fire, so detecting the smoke in time is crucial for preventing fire.From the perspective of smoke detection, this paper will analyse the maincharacteristics of smoke, main work is as follows:Firstly, the domestic and foreign video smoke detection status and common detectionalgorithm are elaborated. At the same time, the characteristics of video smoke detectionare analysed in detail. Master the advantages and disadvantages of various video smokedetection algorithm and present the key points and difficulties of video smoke detection.Secondly, according to the features of smoke vision, a smoke feature extractionmethod based on wavelet texture is proposed. Gaussian Mixture Model(GMM) is appliedfor extracting the motion region. The motion region is processed into blocks andtransformed by two-dimension discrete wavelet to obtain local message. Then, utilizinggray level co-occurrence matrix, each blocks’s textural feature are obtained.Finally, analyze the adaptive neural fuzzy inference system(ANFIS) and put forwardthe united criterion, the combination of them act as the final smoke detection results. Theexperimental results show that the method has a higher recognition rate and lower errorrate than just use the adaptive neural fuzzy inference system.