Study On Flame And Smoke Detection Method Based On Significance In Early Stage Of Fire

Fire seriously threatens people's lives and property safety.If the fire can be detected and warned in time in the early stage of the fire,it will be conducive to the safe evacuation of personnel,control the spread of the fire,and reduce the losses caused by the fire.Therefore,it is of great significance to study the detection methods of flame and smoke in early stage of fire.Video detection technology is the most effective technology to realize early fire detection.In this paper,the method of suspected flame and smoke region segmentation based on visual saliency is studied.The static and dynamic characteristics of flame and smoke are fused,and the additive kernel SVM method is used to recognize video flame and smoke.The main contents of this paper are as follows:(1)Fire suspected area segmentation.In this paper,dimension invariant feature transform(SIFT)algorithm is used to detect flame feature points.Optical flow method is used to extract the optical flow field of moving objects in adjacent frames.LSI color statistical model was used to extract flame color significance graph.Under the bayesian framework,the motion and color significance maps were interactively fused to extract the suspected flame area.(2)Suspected smoke area segmentation.SIFT algorithm was used to detect smoke feature points,and optical flow method was used to extract the optical flow field of moving objects in adjacent frames.Combined with visual significance,motion significance area was extracted,and RGBS color statistical model was used to segment suspected smoke area from motion significance area.(3)Feature extraction of flame and smoke.The flame pulse frequency,texture characteristics based on rotation invariant LBP and GLCM were studied.Based on the temporal and spatial lbp-top operator,the dynamic texture feature extraction method of smoke is proposed.(4)Flame and smoke recognition based on additive kernel SVM.The video libraries of flame and pseudo-flame,smoke and pseudo-smoke samples were established.The extracted features of flame and pseudo-flame,smoke and pseudo-smoke were constructed into feature vectors,and the additive nuclear SVM flame and smoke classifier models were trained.The test samples were used to test the trained SVM flame and smoke classifier.The feasibility and validity of the model were verified.(5)Based on Lab VIEW graphical virtual instrument development platform,flame and smoke detection algorithm is designed.The experimental test of flame and smoke detection algorithm in the early stage of simulated fire verifies the effectiveness of the algorithm.In this paper,an additive kernel SVM pattern recognition method is used to detect flame and smoke,pseudo-flame and pseudo-smoke interference videos in different environments by fusing multi-features of flame and smoke.The results show that the algorithm can effectively identify flame and smoke,and can effectively eliminate pseudo-flame and pseudo-smoke interference.