Research On The Algorithm Of Fire Smoke Detection Based On Video Image

The detecting ability of conventional fire detectors is inherently limited. They often fail to generate timely warnings on unexpected fires, especially in outdoor open areas and spacious indoor areas. In order to solve the problems of fire prevention and fire fighting in the environment of large space, and improve the timeliness and accuracy of fire detection technology, an image type fire smoke detection algorithm based on a combination of various image features of smoke is proposed in this thesis.The algorithm recognizes fire by detecting smoke regions in the image sequence. It has advantages in a few aspects, including a wide range of monitoring, visualization and being non-contact. The research contents of this thesis are summarized as follows:1. Three kinds of moving target detection methods are discussed in this thesis.They are inter-frame difference method, optical flow method and background subtraction method. The advantages and disadvantages of each method are compared and analyzed. The background model is established by using the method of Gaussian Mixture Model and a foreground extraction method based on background subtraction.Testing results show that the method can detect smoke areas simultaneously with other moving targets, and clearly and completely extract the contour of a target area.2. In order to exclude the interference of non smoke target color, which is very different from the smoke color, the moving target area extracted from the motion detection is analyzed by using the color feature of the smoke. And then based on multiple visual characteristics of smoke in images, a few features are extracted by the algorithm to detect the presence of smoke. These features include the salient features,the movement and diffusion features, features associated with the irregularity of smoke regions and those that can make the background unclear. In particular, the shape of profile, the area rate of change, the complexity of the contour and the change ratio of the high frequency energy attenuation are analyzed. The feasibility and effectiveness of each feature extraction algorithm are verified by a number of simulation tests.3. BP neural network is a very useful image classification method, which has the advantages of strong anti?interference ability, low false detection rates and so on.Therefore, this thesis chooses the BP neural network method to fuse and judge the various image features of the smoke, the network structure of BP neural networkwith smoke detection is designed, and then identify whether the image sequence contains smoke regions.4. In order to test the accuracy of this smoke detection algorithm, especially in the presence of interference, a number of image sequences which are taken in indoor and outdoor environment are selected as test objects. The test objects include both smoke images and images that do not contain smoke. Test results show that the algorithm can accurately and quickly detect fire smoke with a low false alarm rate and high reliability.