Automatic Fire Detection by Changing Region Analysis Based on Geographic Information

This paper proposes a new method for automatic fire detection in cities using image processing techniques. Images are obtained successively at regular intervals from monochrome TV cameras which are suitably fixed on roofs of tall buildings. Fire detection is camed out by identifying smoke in the daytime and fire flames at night. The major feature of the proposed method is to employ a fire identification strategy which is applicable to both smoke and fire flames and efficiently uses geographic information for identifying fire location. The strategy is as follows; changing regions are extracted as fire region candidates by image subtraction operation. Motion analysis based on changing region tracking is camed out. As a result, actual fires are discriminated from disturbing events such as swaying trees, moving cars, etc. It is thought that smoke or fire flames might be detected in fragments due to noise. Fire spreading inference based on geographic information is adopted for top down grouping in changing region tracking. Experimental results have shown the effectiveness of the proposed method. The Geographic information systems (CIS) have come into practical use in many fields, for example, utility (water, gas, electric power) supply management, urban planning, etc.[l]. The main reason for the wide use of CIS is the rapid improvement in computer capability, such as processing power and storage data capacity. By the way, interest in the prevention of disasters has been increasing recently in Japan due to fear that a big earthquake will happen. Several projects on urban disaster management systems as an application of CIS have been progressing by governmental organs. One of the problems in an urban disaster management system is how to manage fires which simultaneously break out at multiple locations after a big earthquake. This paper proposes a new method for automatic fire detection in cities as a key technique for the urban disaster management system. Fire detection is camed out by identifying smoke in the daytime and fire flames at night from images in the visible band. Fire location is calculated by matching a part of the image with an area code described in a geographic database. Roughly speaking, there are two ways for fire detection. That is, from images in the visible band using ordinary ITV cameras and from images in the infra-red band using thermal sensors or similar devices. The former has an advantage from the viewpoint of cost and maintenance. There are several studies on fire detection using thermal sensors. However, reports on automatic fire detection which deal with both smoke or fire flames in the visible band are very few [2]. Furthermore, they do not discuss about the problem of fire location identification. The major feature of the proposed method is to employ a fire identification strategy which is applicable to both smoke and fire flames and efficiently uses geographic information for identifying fire locations. The basic idea of the proposed method is described in Section 2. The fire detection method is explained in detail in Section 3. Some preliminary experimental results are show in Section 4.