Research On The Fire Flame Recognition Based On Image Processing

With the progress of society, the number of the high-rise building is growing, so the fire safety has attracted more and more attention of society and it is particularly important to detect and control the fire at initial stage as soon as possible. The signals of traditional fire detection technology such as smoke type, light type, temperature-based type, etc. become very weak because they have some disadvantages such as smoke is easy to spread, light source acquisition is vulnerable to outside interference and the temperature can’t pass quickly etc. However, the image processing methods can be used in the research on fire monitoring based on image processing to extract fire image at the initial stage of fire for monitoring and detecting fire rapidly.The mainly purpose of the fire’ flame detection and algorithm which based on digital image processing is to make sure whether the fire is existing. The fire images’ are got by high resolution capability CCD camera. These images will be sent to the computer. By choosing proper image processing algorithm, we can distinguish the existing of fire from one image or series frame of images. This paper will talk about how to achieve the fire detection which based on the fire’s color features, grayscale features and shape features. The image digital processing is the necessary knowledge for this study. The purpose of design is to choose right algorithm and select available ways to detect and identify the fire.A fire flame recognition algorithm based on image processing is presented in this paper. First, the suspicious targets of the input image are extracted by combining the method of background subtraction with flame color features, then after image morphological processing, with the regions labeling in consequence, circular degree values of region are calculated to filter out some interference regions whose circular degree are bigger. Secondly, the dynamic features of fire image are showed in this paper, along with the similarity of flame images, the variation of flame pixel grayscale value and the variation of flame fire angles’ number, at the same time, an algorithm of fire angle detection is proposed according to the four features of angle. It is found in the experiment data that the three characteristics can effectively distinguish between fire and the interferences, and a proposed use of these three dynamic video features as parameters to identify a fire is described based on the above. Finally, the basic theory of BP artificial neural network is described, as making these three dynamic video features as neural network input vectors, BP artificial neural network is designed to identify the fire flames.The experimental results show that the fire detection algorithm with mixture of multi-criteria can identify the real fire and have a good anti-interference ability and fire detection based on image processing has a good development prospect.