| The technology of forest fire detection based on video analysis is an important method of Forest-fire prevention.Compare with traditional technology of fire detection,the technology of Fire-detection based on video image analysis has many advantages.In the early stages of the fire,smoke is easily captured by the monitored video while the flame is easy to be blocked.Therefore,the forest fire detection technology based on video image analysis is essentially to detect the fire smoke image in the video.In this thesis,to detect the forest fire smoke in video,studies the problem of moving object extraction,image feature extraction and classifier design.The main research contents are as follows:(1)Research on moving objects extraction algorithm in video images.This thesis focus on the GMM and the ViBe.In order to compare the performance of the moving objects extraction algorithm,we introduces a method to evaluate the moving objects extraction algorithm by using the method which classifies the image pixel set in this thesis.To divide the target image from the video image,this thesis study the method of using Freeman chain code to express contour and obtain the external rectangle of the contour.(2)Research on image feature extraction algorithm.We study a variety of feature extraction algorithms contains LBP,GLCM,LBP-TOP,wavelet energy feature,color histogram,color moment,motion direction feature and compactness.This thesis introduces a method to search the optimal feature subset by using KNN feature select algorithm.(3)Research on classifier design.The experimental results show the classification of the features or features combination.The optimal parameters of SVM are obtained by grid search method.By comparing with BP-NN,verified the superiority of SVM in the duty of smoke detection.(4)Design and implement a software which detect the forest fire smoke in video image.Face the research on forest fire smoke detection algorithm based on video analysis,designed and implemented a software which can detect the forest fire smoke in video. |
