Research On Fire Prewarning Method Based On Video Content Analysis

Fire is one of the most dangerous disasters that seriously threaten human life and property. At present, the conventional fire detectors used in our society mainly rely upon temperature sensors or smoke sensors. In real applications, the reliability of these fire detectors always suffer greatly from some adverse environmental factors, such as distance, dust, humidity and so on.In recent years, with the performance improvement of embedded processors and emergence of dedicated graphics acceleration chips, video-based flame detection technology has gradually attracted more and more researchers. The video-based fire detection technology combines computer vision, digital image processing and pattern recognition together for fire pre-warning, which can automatically identify fire incident evolving in relatively complicated scenes.According to the physical features of fire at early stage, a per-warning system of dual-band video-based fire detection is presented in this thesis. The proposed algorithm contains three parts: object region segment, feature analysis and object recognition. Firstly, considering the different characteristics of flames in infrared images and visible images, different segmentation methods are adopted. In infrared images, the candidate flame areas are detected by using the region growing method. And in the visible image they are obtained by using the color and motion information of fire. Then, the static and dynamic characteristics are combined to describe visual features of flames. Finally, the real fire regions are detected by using a synthetic criterion from dual-band images.The experimental results show that the proposed algorithm can detect fire states effectively with high accuracy.