Fire Detection Research Based On UAV

In this paper fire detection technology on UAV platform is researched, the practical significance of this research is analyzed, domestic and international research status of fire detection is reported, and a high accuracy, low false positive fire detection system is designed, which is able to identify the early fire accident and position the fire.The system uses the embedded microprocessor and Linux operating system as a platform, catches the infrared and visual characteristics of the fire flame with an infrared camera and a color camera carried on UAV. Video for Linux Two, the application programming interface of the Linux operating system, is used to acquire images. The collected images are transmitted through corresponding network transmission protocol. The images obtained after transmission are processed to complete the fire identification work and to give the position information of the fire. Then a turntable fire tracking system can be used to track the fire.In this paper, the static and dynamic characteristics of fire flames are analyzed in detail, such as the characteristics of brightness, color, shape, area variations, edge changes, physical variations, flashing and so on, base on which the corresponding image processing algorithms are presented to detect the early fire flame. Firstly, the threshold segmentation algorithm is used in the infrared images by the high temperature to get the binary images. Then, closed operation is adopted to connect the suspicious target area. After that, the suspected high temperature flame zone is got. Secondly, the color segmentation algorithm is used to extract the suspected high temperature flame regions, which are flame colored. These areas are called suspected fire areas. Thirdly, the area growth algorithm is used to extract the target area, whose area is grown trended. These regions can be called highly risk suspected fire areas. Fourthly, the recognition algorithm based on the changing characteristics of contours is adopted to extract the high variation profile, excluding the smaller amplitude profile. Finally, the results of above four algorithms should be the inputs for data fusion. After data fusion the final recognition results can be obtained. Experimental results show that these algorithms have higher accuracy and noise immunity, lower false positive rate, can accurately identify the fire flame, effectively filter light, stable candle flame such interferences.