Research on Image Recognition Algorithm in the Forest Fire Prevention System
|School||Harbin Institute of Technology|
|Course||Information and Communication Engineering|
|Keywords||flame recognition smoke recognition image segmentation dynamic character|
Forest is the main terrestrial ecosystem, with high ecological benefits and economic benefits. In view of the current forest fire in China’s serious situation, effective technology must be develop to solve the problem of forest fire monitoring so that people’s lives and property safety can be effectively protected. More traditional fire detectors use a single moment of the parameters as a standard, in the under-interference caused frequent false positives or omissions. In recent years the fire alarm system based on the machine vision uses digital image processing techniques to achieve automatic fire alarm.Based on the flame and smoke image characteristics, a machine vision method of identifying the natural fire is proposed in this thesis. In the course of the fire, the main image information is the combustion of smoke and flame. Through the study of smoke and flames image information, smoke and flame phenomenon itself has certain regularity. So targeted algorithm can be designed, identify the smoke and flame from image and judge whether the fire occurred based on this kind of found.First of all the thesis explains the techniques status and development of forest fire prevention and fire-detection using digital image processing techniques. On this basis, the segmentation and the identification of flame and smoke are discussed.Then, three different segmentation technologies for different flames are proposed to achieve the accuracy flame region. For the detection of the flame characteristics, color and dynamic analysis are mainly used. The color is identified by establishing of flame color model. Further the dynamic characteristics of the flame are identified. For the complexity of the smoke color, the color extraction method is used to division, and improved by using of the clustering algorithm for visual consistency. For the detection of smoke characteristics, the wavelet characteristics analysis and the dynamic characteristics analysis are mainly used. The wavelet characteristics are identified by comparing images and background images smoke wavelet coefficients, and then the dynamic characteristics of the result, including the irregularity and the diffusivity of the smoke is further identified. And then we can determine whether there is smoke in the video.Finally, the overall flow of fire identification in forest fire protection is proposed based on above analysis.The experimentation results show that the fire detecting method which integrating static character and dynamic character of flame smoke has high recognition rate. In the area of fire detecting based on video image sequence analysis, the technique introduced in this thesis has good prospect for development .