Study On The Algorithm Of Indoor Scene Flame Detection And Design Of Visual Alarming System

Fire has been threatening the safety of human life and property since ancient times.It is an important research topic to detect fire in advance and then issue alarming information.For nearly two decades,with the rapid development of science and technology,fire detection technology based on vision has become a major research direction to prevent fire.Visual-based fire detection using surveillance video can get richer information,more anti-interference ability,and the application of the scene is also more extensive.However,the current visual-based fire detection research work is mainly for the burning flame,and no scholars have studied the initial flame.In the bus,gas station,flammable warehouse and other scenes,once initial flame appears,it may cause immeasurable losses.According to the needs of these special scenes,this thesis studies the static characteristics and dynamic characteristics of initial flame for the indoor scene of flame videos,and puts forward a complete set of flames detection method.Through the flame videos test of different scenes,the method proposed in this thesis has a lower false alarm rate and higher accuracy rate.The main contents of this thesis are as follows:(1)The color characteristics and contour shape characteristics of initial static flames are studied.And the statistical characteristics of flames under different color models are analyzed.The flame image is segmented by the color characteristics of the flames.Then results are compared and analyzed under different color models to get the best color model that represents the flames.The Fourier descriptor and Hu invariant moments are used to express the boundary shape of the flames.And the reliability of the Fourier descriptor and Hu invariant moments are compared and analyzed.(2)The motion target detection in flue video is studied.The moving target in the flames video has its unique movement characteristics.When the flames appear,they can be regarded as moving objects.After the flames appear,they are still stationary.Aiming at the characteristics of flare movement,an improved frame difference method is proposed to detect moving objects,and a lot of background information closed to the color of flames is removed.(3)The dynamic characteristics of flames between adjacent frames are studied.Flame is a target with a stable shape and continuity between the frames.The flames have a correlation at the coordinate positions between two adjacent frames.By comparing the position information between frames,the correctness of flames detection can be verified.(4)Support vector machine is applied to determine whether the flames exist.A large number of flame samples and negative sample images are obtained by using the flames of video in different scenes.The support vector machine is used to train the positive and negative samples.Then the accuracy of the flames is tested by the test samples,and the test results indicate that the method has a higher recognition rate.