Research On Video Based Fire Smoke Detection Method

In recent years, there are many fire accidents especially in open area and large-scale indoor places. In these places, traditional fire detectors are incapable of reporting fire accidents in time which leads to great losses. As the development of computer and machine vision technology, a brand new fire detection system called video-based fire detection system has emerged in the past few years. This type of detection system uses cameras to capture image sequences of the monitored area, pass the image sequences to the computer and process the images using machine vision methods to decide whether there is fire in the monitored area. Because the camera has wide field of view in open area, the system is especially suitable for fire detecting tasks in open area.There are two typical types of fire detector: fire-based detector and smoke-based detector. Because smoke is visible in the early-stage of fire, smoke-based detector is better for early-stage fire alarm, so we mainly studied smoke detection in this thesis. The thesis is organized as follow:Firstly, we studied motion area segmentation methods, described background subtraction method and Gaussian mixture background modeling method. We also studied connected areas extraction and marking methods so that we can locate the suspicious smoke area rapidly and get prepared to track the moving target.Secondly, we analyzed the image area extracted by our suspicious smoke area extraction method. We studied the generating process of fire and the visual features reflecting the physical characters of fire like smoke point stability feature, smoke area expansion feature and smoke diffusion direction feature.Thirdly, we studied the static features of smoke region. We described the gray-scale consensus method and studied the texture consensus feature of the smoke region using Local Binary Pattern. Our experiments showed that classifying smoke using LBP method is possible. Lastly we analyzed the structure of the existing video-based smoke detection systems and proposed the improvements such as using existing surveillance equipment to minimize system cost. We also studied the methods of combing multiple smoke features to achieve precise and robust smoke detection.