%A XIE Di, TONG Ruo-feng, TANG Min, FENG Yang %T Distinguishable method for video fire detection %0 Journal Article %D 2012 %J Journal of ZheJiang University (Engineering Science) %R 10.3785/j.issn.1008-973X.2012.04.018 %P 698-704 %V 46 %N 4 %U {https://www.zjujournals.com/eng/CN/abstract/article_38370.shtml} %8 2012-05-02 %X

A novel method for video fire detection based on artificial neural network was proposed in order to estimate fire regions and predict occurrence of fire in video surveillance system. Temporal-spatial features were analyzed including flicker frequency and geometry of fire by Fourier transformation, roundness degree analysis and corner detection based on motion and three-dimensional color features. Then these features were used as a probability vector fed for artificial neural network classification model to output the probability of fire. With keeping accurate detection rate, two problems, local extremum tendency and slow convergence rate, of artificial neural network were solved by optimal parameters combination experimentally. The method can distinguish flickering vehicle light from actual fire in varied locations, e.g. tunnel, warehouse and museum, to prevent ambiguity and eliminate the influence environment illumination imposing onto the detection results which significantly reduce false positive rate. Experimental results show that the method has a 96% detection rate while keeping a real-time performance.