Research On Flame Detection Algorithm Based On Image

With the development of social economy and the progress of technology, especially the sharp increase of the urban population and the fast developmentof urbanization, more and more higher buildings have been built and large-space places emerged .In those higher buildings fire could spread rapidly; moreover, the fire-fighting and rescue work will be very hard to conduct. Therefore, how to effectively protect the safety of high building and to detect the possible fire accident as early as possible has been an increasing urgency. As a new effective measure to detect the possible fire in an early stage, the fire-detection technology based on image has raised many concerns. In this paper, the digital image processing techniques have been discussed and several flame feature extraction algorithms and the fire detection systems based on images with the BP neural network have been designed.The algorithms of fire image enhancement, noise filtering, image segmentation, motion detection, fire target extraction and the applications of these algorithms in fire image processing have been studied in this paper. As a result, the fire target can be extracted from the fire image sequence through integrating several segmenting algorithms.In this paper, the flame features are extracted from three aspects: first, the profile of flame is extracted based on the color feature of fire; second, the flame images are extracted from the static features of fire, basically including the moment feature, curvature feature, fractal dimension and so on; third, the flame images are extracted from the dynamic features, mainly including the features of increasing fire area, similar shape of fire, the overall movement of fire and so on. A description method of sharp corner of fire which is based on curvature and the extraction method of fire profile with the color features of fire are studies therein. Further, the description of fire profile with fractal dimension is also studied. Especially, some improvements about the algorithms of the curvature feature are made and extract the profile of the flame used color feature. From these features, the early-stage fire image can be detected while some interfering phenomena can be distinguished. These fire features can be further detected with the neural network.The fire detecting scheme based on neural network has been discussed in the last part of this paper. First, the basic concept of neural network has been introduced. Second, the detailed structure of the BP nerve network and the detailed design of the input and output layers have been concluded. On these bases, the fire detection system based on BP neural network has been designed and experiments with a series of fire image samples and interference mode images have been conducted. The experiment results show that the BPNN fire detection system has strong capacity of anti-inteference.