Investigation Of The Fire Detection Technology In Freeway Tunnel Based On Visual Saliency

Fire is one of the most serious accident that may cause high casualty happened in the freeway tunnel. Recently, due to the construction of the freeway in mountain area the freeway tunnel mileage number increases as well. Nowadays, the fire detection in freeway tunnel has been a main challenge and agent task. The conventional fire detection approach is based on the individual detector. It shows disadvantages such as low discrimination, time consuming judgment procedure and high maintenance cost. Therefore, to detect the fire in time then take step to reduce death and property loss, we proposed a novel fire detection method which is based on the image obtained by the installed surveillance system. An image processing technique was applied to identify the flame in the surveillance image.The visual saliency detection can provide a method to identify the objects with high attention rate among other objects and flame just shows such high attention rate property in the freeway tunnel. Therefore, in this thesis, visual saliency detection was adopted to detect the fire in freeway tunnel.In this thesis, we introduced the basic knowledge on video image processing technique applied to detect the flame. Based on which, the surveillance video was remodeled to image series, then the moving objects in the adjacent images were extracted and described by the optical flow approach. Using this method, the fake signal that may case by the immobile light in tunnel can be eliminated. After the moving objects were captured, the superpixel segmentation was exerted on the surveillance images using the SLIC method. Using the superpixel as the arithmetic element, the saliency computing process can be simplified. The optical flow information was extracted when the saliency computing process was conducting and the histogram of the optical flow information was inputted as the visual saliency. Unlike the conventional computing method which only detects the object or background of the image, the approach proposed in this thesis detects the object and the background simultaneously by constructing a two layer visual saliency model. For the first layer, based on the image distribution properties, the visual saliency computing was conducted based on images with the superpixel in the up, down, left and right boundaries as the background marking point. Then computing results based on the four different images were merged and regarded as the first layer visual saliency computing result. Then the adaptive threshold method was adopted to obtain the visual saliency computing result of the second layer. Finally, the results of the two layers were merged to detect the fire. This method provides us a more accurate way to detect the fire.This thesis proposed a novel visual saliency method to detect the fire in freeway tunnel based on the surveillance video. This method can eliminate the interference resulting from the light and automobile tail light. It shows good real-time performance and strong capacity of resisting disturbance and can be applied for the fire detection in the freeway tunnel.