Research On Highway Tunnel Fire Flame Detection Method Based On Image

Fire is the worst kind of safety accidents in highway tunnel except for the traffic accidents. Because of the special environment of highway tunnel, in case of fire, not only be vehicles and personnel evacuation difficult, but fire rescue is harder in a short period of time, which often can cause significant casualties and property loss. In the early stage of the fire, if the fire can be alarmed accurately and timely, it can be suppressed in time to reduce the loss caused by fire. Therefore, it is important for early detection of tunnel fire. AdaBoost pattern recognition method and the image characteristics of flame are used to detect the flame in video combined with the special environment of highway tunnel in this paper. The main contents of this paper include the following parts:(1)Moving object detection. Respectively, using the inter frame difference method, the Gaussian mixture model background subtraction method and the optical flow method to extract the motion region, in this paper, the inter frame difference method is selected to extract moving object region by analyzing and contrasting the characteristics of the three methods. And then the motion object images are post-processed.(2) Candidate flame region segmentation. In this paper, an improved method for segmentation of suspected flame area is proposed, namely the segmentation method of combining RGB with Lab color space. The experimental results show that the improved segmentation method is effective.(3) To detect the characteristics of flame. In this paper, there are six flame features, namely the first order moment of the H component, the ratio of a, b component in Lab color space, the rectangularity, the circularity, the sharp angles and jumping frequency. And the distribution rang of flame and interference characteristics data is analyzed by experiment in this paper.(4) Flame recognition based on AdaBoost. The sample data extracted from the flame and the interference video forms the feature vector and is input into SVM and AdaBoost, forming training model. And the parameters of SVM and AdaBoost are optimized to improve recognition accuracy. By analyzing and comparing the recognition accuracy of SVM and AdaBoost, AdaBoost with the higher detection accuracy is selected as the recognition method in this paper.AdaBoost pattern recognition method and flame characteristics are used to detect the flame and interference video. The experimental results show that the recognition method of this paper can effectively detect the flame in video and eliminate the interference of false flame automotive lamp in the video of highway tunnel.