Research Of Essential Technology In Video-Based Early Fire Detection For Large-span Structures

With the rapid development of social economy, the rescue in disasters of large-span buildings is taken more and more attention in urban safety management. As one core component of it, early fire detection has a broad application prospect in many aspects, such as fire surveillance, fire warning and responsibility analysis. At present, Early Fire Detection is focused at home and abroad.Given to the influence of complex environment in the surveillance scene and multiple interfering factors, Early Fire Detection has always been one of the most difficult problems in the field of pattern recognition for a long time. In the thesis, based on the in-depth study to Early Fire Detection, one set of effective solutions is proposed and an actual software system is developed. The research work and achievements of this thesis are as follows:(1)Initial input video images are processed by a variety of image preprocessing techniques, while fire flame is recognized as a special moving target by using background subtraction. A method of median value with adaptive updating mechanism is proposed to establish the background of region. This method can establish the background precisely and rapidly, and the background will be reestablished as soon as the mutation takes place. All these above lay the foundation for detecting the moving target and extracting its characteristics;(2)The feature of early fire flame are studied and discussed respectively. The integrated feature of Illumination—Saturation of Color Predominant Component, the feature of area changing and the integrated feature of Edge—Centroid are proposed. Based on color space analysis and Contourlet Transform all the above features have been extracted.;(3)Based on the analysis of natures of fire flame itself, an algorithm which combines method of threshold with algorithm of BP neural network has been proposed. The algorithm divides the process of target recognition into two parts, which represent both two features. The algorithm can fulfill the requirement of accuracy and speed at the same time;(4)The algorithm proposed in this thesis has been developed and simulated with Matlab 9.0, and the comprehensive early fire detection software has been designed. The results of experiment show that the accuracy can reach 94%, with the average detection period of only 10 seconds, which meet the requirement of design.Based on the existing samples of early fire flame videos, the method proposed in this thesis has been proven to be advanced, effective and practical via experiments. The system designed has good characteristics of real-time and accuracy, along with good robustness and adaptability.