Image Flame Recognition Based On AdaBoost And SVM

Fire detection based on video image has the advantages of fast reaction, wide rangeof detection. So it can effectively solve the failure of early warning problem in the largespace building environment happened on the traditional fire detector. Thus, it has abroad application prospect.Image flame recognition technology type including image segmentation, featureextraction and recognition of the three links. This article outlines the principles andfeatures of video flame detection, focusing on image-based flame detection algorithmbased on AdaBoost and support vector machine combining called AdaBoost-SVM. Firstto frame the fire video, then using background subtraction combination of fuzzy integralmethod extract the moving targets of video sequences and then extracted suspectedflame region by using the criterion of flame color，Analyze and calculate the fourcharacteristics-area change rate, circularity, correlation coefficient and area ratio ofsuspected flame area, as input parameters of SVM, using AdaBoost algorithm to markthe SVM samples which are misclassified mainly, increase the ratio of the misclassifiedsample, so that it can be re-elected more likely. Continuously iterative training focusedon "the most information of sample points", In order to solve the problem in traditionalAdaBoost algorithm which only considers the training set weight of the base classifier,without considering the effects of under test samples for weight problems, it willcalculate the distance between sample under test and the fault sample points of eachround, adjust the weight dynamically according to error rate effect on baseclassifier,finally get the final classification result from all the base classifier usingvoting mechanism.,compare the LibSVM and AdaBoost-SVM algorithm through thesimulation experiment. The result shows that the AdaBoost-SVM can effectively improve the precision of data classification and has high recognition accuracy.