The Research On Forest Fire Smoke Detection Algorithm Based On Feature Fusion

The smoke detection technology based on the video is a major breakthrough in the traditional forest fire detection system. Based on the analysis of present situation of the smoke detection, it gives a smoke detection algorithms,combines moving object detection, feature extraction with classification identification.Firstly, in order to prevent the interference of light disturbance and branches mutation, method in improving both the median filter and gaussian mixture model respectively are proposed in this article based on the background difference method. Method 1 is to improve the median filtering method to extract the background model, then motion area is extracted by difference image campaign cumulative discriminant; Method 2 combines bilateral filtering and gaussian mixture model to segment movement area, and then connects domain processing to extracted results, so that eliminating the interference of distractors further.Secondly, the detected moving area is feature extracted to adapt to the impacts of complicated background conditions of forest fires and weather conditions and other factors, this paper presents a multi-feature fusion algorithms, extracts color feature, texture feature, centroid distance features and HOG characteristics orderly, and these features consist of a large feature classification to be used by the input matrix. HOG histogram is an algorithm recently proposed which is characterized by an image processing, and appearance and shape features can be characterized because this feature geometric invariance and local contrast changes are not sensitive features, so the article introduces smoke detection system, and the experiment proved that feature extraction is operable to smoke and has a good effect.Finally, the existing classifier is introduced, especially the support vector machine and least squares support vector machine, and least squares support vector machine has better performance than the support vector machine by experiment. Experimental results show that the proposed algorithm has better recognition results and improve the accuracy of detection.