Early Fire Alarm Based On Machine Vision Method

Fire is one kind of the frequently happen natural calamities, which seriously threat to human being's life and society safety. Studying the law of fire's occurrence and development, especially studying the detection method of early phase of fire have most important significance to protect the people life and property's safety.The traditional fire probes often use one moment fire parameters as the judge standard, In the process of probing, react slowly, difficult accurate alarm in the fire early phase, frequently do the wrong warning or omits warning under the external environment's disturbance. Especially in the building that is big space, or full of dust, or high humidity, the traditional fire probes hardly can work regularly.In recent years, fire alarm system based on machine vision is brought forward, the system realize fire auto alarm using digital image processing technique. In the fire developing processes, the main image information is smoke and flame coming into being during burning. By studying the image information of smoke and the flame, we found the smoke and the flame phenomenon have certain law, take the law as basis design algorithm and detect smoke and the flame from the image, judging if a fire has happened.This thesis proposed a method which giving early warning to fire based on machine vision. By analyzing and processing images captured from the camera, making use of color and dynamic character to detect the fire. To smoke detection, mainly analyses the color and dynamic character, building the smoke color mixture Gaussian model to detect color, the result of the color detection is the basis of the dynamic character recognition. The dynamic character includes anomaly and diffusion. Finally, confirm whether the fire occur in the video scene. To smoke detection, mainly analyses the color and dynamic character, building the flame color mixture Gaussian model to detect color, the result of the color detection is the basis of the dynamic character recognition. Aiming at flame area ununiformity brightness, have realized flame area brightness space analysis. Aiming at the flicker of the flame,three detection algorithms are designed: 1) detection algorithm on the brightness of flame pixels; 2) detection algorithm on the frequency of twinkling; 3) analysis of disorder in the flame region. In the view of the continuing state of flame, continuing detection algorithm is designed; in the view of the growth of the flame region, an increasing algorithm is designed to detect and to get the region of the flame. To get more precise flame region, the detected region is raised, and judged by color, growing the near region of to the flame pixel, and in the end, more accurate flame region has been got.The experimentation results show that the fire detecting method which integrating smoke's and flame's static character and dynamic character has high recognition rate. In the area of fire detecting based on video image sequence analysis, the technique introduced in this thesis has the good development prospect.