Research Of Fire Detection System Based On Infrared Image In Cabin

Fire accidents in the ship usually caused serious casualties and property losses. In order to ensure the safety of life and property on ship, effective fire detection method for the safety of ship cabin need to be researched.Restricted by the ship environment, traditional fire detection system based on sensor fails easily, and it has high rate of false positives. The fire situation based on video image monitoring in the scene has the advantages of fast reponse and is not restricted by distance and space environment, the study of new fire detection method has important practical significance. At the beginning of the fire accident, smoke is usually the common phenomenon, by analyzing the characteristic of smoke image information, some fixed features of fire smoke at the beginning of the fire were obtained, which were used as the basis of recognition of fire smoke to determine whether a fire accident in the video scene has happened or not.First of all, according to the characteristics of ship cabin environment, in this paper, a method used to detect suspicious smoke area based on template separating and matching method was proposed. Firstly, the visible light images and infrared images were distinguished by RGB space model method. Secondly, the acquired images should be pretreated by methods, such as graying, denoising and blocking. Finally, the suspicious smoke region could be detected by template matching algorithm. According to the changes of the splits’number of ei (Template Matching Value), a suspicious smoke weather exist could be judged.Secondly, after determining the suspicious smoke area, the advantages and disadvantages of optical flow, inter-frame difference, background subtraction and background subtraction based Gaussian model motion region extraction method were compared from the segmentation effect and the feasibility of running performance points of view. Then, Gaussian model difference method was chosen to extract suspicious smoke area.Then, three features including the growing of he area, circularity and immovable position of smoke source spot were extracted as the main judgments of fire detection. Then three characteristics value of smoke were inputted into the trained Bayesian model, and fused with Bayesian decision to determine whether there is a fire in the scene.Finally, in order to verify the validity of the algorithm, experiments in different scenarios and different periods of fire with video and without video were tested, the test results showed that the developed fire detection algorithm can effectively detect fire scene, and has good anti-interference ability.