Research On Fire Smoke Detection Algorithm Based On Video Sequence In Urban Garden

With the continuous development of science and technology, urban modernization construction has made great progress, in order to create a healthy, harmonious, green, open, sharing of the new city, city park, more and more appear in people’s vision. The city park not only to improve the people’s quality of life and promoting ecological harmony and of windbreak and sand fixation, beautify the environment, purify the air, reduce noise, water and soil conservation and has a prominent contribution. But frequent garden fire is a serious threat to the safety of people’s lives and property, to a large extent hindered the development of urban, so the prevention and control of the urban landscape and scenic fire has been more and more extensive attention. In view of the above problems, this paper on the existing domestic and foreign video fire detection technology were comparison and analysis and neural network technology and the digital image processing technology combined, the fire smoke detection method based on video image processing, the main research contents of this paper package includes the following several aspects:(1) Combined with the characteristics of smoke based on fuzzy characteristics and testing environment variable. In this paper, the based on Kalman filtering of background subtraction method to extract the suspected smoke area; and the introduction of corrosion and expansion of two operators of mathematics morphology of slight noise in the image processing.(2) Through the analysis of the characteristics of domestic video smoke extraction technology, select the static and dynamic features with a certain discrimination as input features. Static features are as follows:according to color features, through analysis of smoke and ash in the RGB space model characteristics, smoke color decision conditions are determined, and the suspected smoke area represented by gray pixels, while the introduction of pixel ratio to characterize the color; for texture feature, the paper uses the gray co-occurrence matrix method is used to extract texture features of the image, and choose a representative of the second order moments are used to describe the texture feature.(3) The dynamic characteristics of the research are as follows:according to the fuzzy feature, the method used in this paper to construct the two-dimensional discrete wavelet transform smoke background fuzzy model, and introduces the attenuation of high frequency characteristics of fuzzy rate characterization of objects; for irregular shape features, this paper uses Canny operator to extract moving regions of the boundary shape, and according to the extraction the perimeter and the motion region boundary chain code method and scanning area, finally introducing a circular motion region to characterize the degree of irregular shape characteristics; for the main direction of movement of smoke characteristics, by comparing the commonly used block matching algorithm, select the three step (TSS) with the minimum absolute value and function block matching, and calculate the moving direction of each block through the discrete motion direction search method, finally introduced the movement direction of the main transport ratio coefficient of objects Moving direction.(4) Design based on BP neural network image fire intelligent recognition system structure, and color feature (color pixel ratio) and fuzzy characteristics (high frequency energy attenuation rate), boundary shape irregular features (roundness), texture (angular second moment) and the main movement direction feature (movement direction of proportion coefficient) as the input signal for feature fusion and test of different networks and their shooting scenes video smoke and non smoke video simultaneously.The experimental results show that the video smoke detection algorithm in this paper has good real-time performance, can quickly and accurately detect the smoke in video, effectively reduces the false alarm rate.