The Design And Validation Of The Fire Detection Algorithm For Self-service Bank

As there are few people in the self-service halls of China bank in the midnight,some criminals will deliberate arson and take other criminal activities in thesepublic and unattended scenes. When a fire occurrs, it often causes great economiclosses and casualties. Therefore, for self-service halls of China bank, we shouldenable early detection of the fire, allowing more time to respond and extinguishfires to avoid a major disaster. To meet the needs of the bank’s video surveillancesystem, this paper proposes a fire detection algorithm with the goal to detect the fireat the earliest stage.This dissertation mainly studies how to extract the possible fireregion, and proposes a weighted fusion method of multiple features for firedetection. We also conduct considerable experiments to test the fire detectionalgorithm. The main content of this dissertation are as follows:(1) The thesis designs a possible flame region extraction algorithm based onthe flame movement and color characteristics. As the burning flame jumps, we useMotion History Image(MHI) method to find the foreground region of the currentimage. Because the color of the flame is quite different from the surroundingenvironment. we study the color characteristics of the fire in the RGB color space,HSI color space, and YCbCr color space, respectively. In the dissertation, we usethe motion characteristics and the color information of the fire to extract possibleflame regions. The experimental results show that we can effectively identify theflame area.(2) The thesis proposes a weighted fusion algotithm of multiple features forfire detection. This thesis proposes the fire characteristics weighted fusion algotithmto identify the flame regions by analyzing of a large number of real fire and non-firevideos. The main characteristics exploited by the algorithm are as follows: thetexture characteristics, the measurement of disorder, variation of the location, thechange rate of the fire region, the temporal wavelet feature, and the saturation offire.(3) We conduct many experiments to test and analyze the fire detection algorithm. In this thesis, we apply the proposed algotithm to the fire detectionsystem for self-service banks. A lot of video analysis and verification shows that thesystem performs very well.