| Fire on board is one of the most dangerous disasters at sea as it can damage the cargos, can cause water and air pollutions and even can cause loss of lives. Although there are many ways to prevent or fight against fire, there have been many fire accidents on board in recent years.The main purpose of this research is to investigate the feasibility of digital video-based vessel fire detection system which is more robust, reliable and capable of flame recognition even in the camera shaking conditions that are unavoidable in rough seas. Besides, the system is to be implemented for the detection of fire flame in more distant places. In order to accomplish the research objectives, the hardware equipments including personal computer (PC), Universal Serial Bus (USB) video camera, fire ignition box as well as the software tools including Visual C++ programming, DirectShow and OpenCV are used.The digital video-based fire detection systems have many advantages over the traditional fire detection systems such as low equipment cost, faster response, being capable of monitoring a large area and being able to confirm the existence of a fire through the surveillance monitor without visiting the location. Based on the fire information detected by such system, the most effective fire fighting method can be chosen and applied properly to extinguish the fire.The tested fire flame is mainly light blue in color. Except the white color flame, the blue flame which is produced from the complete combustion due to the pre-mixing of sufficient oxygen and fuel is the hottest and it has more heat energy than any other colors such as red and yellow in a flame. For example, the temperature of more bluish oxyacetylene flame is about 3000℃while that of the candle flame is about 1400℃. Early detection of the more dangerous light blue flame is essential for choosing a proper fire fighting method, and hence, the light blue flame detection is mainly carried out in this research. The program in this research is divided into three parts:flame recognition algorithms, anti-shake filtering and real time video transmission over LAN.For the flame image recognition, two flame detection algorithms:split block algorithm and region of interest based (ROI-based) color intensity composition detection algorithm are proposed. Both algorithms can distinguish well between the fire and non-fire objects as the detection of fire character pixels is considered under the four different background lights:dark, dim, normal and bright. Compared with the split block algorithm, the ROI-based fire detection method save the program execution time since it needs to check only within the selected ROIs and it doesn't check the whole image for fire character pixels.If the digital video-based fire detection system is used on board, shakings of the camera or scene are usually unavoidable because the ship is moving in dynamic sea states such as waves, wind, current and tide or inclement weather conditions. In case a motion blurred scene image is detected by the digital video-based fire detection system, the undesirable false alarms will be raised. In order to prevent or reduce the false alarms, image blurring effects must be removed from the detected images. Therefore, image deblurring filters must be properly applied in the image pre-processing stage of fire detection before analyzing the scene image for fire character pixels in the image processing stage of those vision-based vessel fire detection systems. To restore a motion blurred image to its near original form, the motion blurring parameters (blur length and blur angle) of its point spread function (PSF) must be properly estimated. In this research, new estimation methods of PSF parameters are proposed. In order to get the best PSF, the motion blur length and blur angle estimations are made by using the Discrete Fourier Transform (DFT) spectrum.For the estimation of the motion blur length, two new simple approaches are proposed. In the first approach, the image is first rotated counterclockwise according to the inclination angle of the DFT spectrum stripes. Then the maximum pixel value in each column of the image is checked vertically from top to bottom of the image. In order to reduce the iteration time, the central part of the image that is the most pixel intensity concentration area is selected. The second approach is considered without rotating the spectrum matrix. According to the results, the second algorithm is better than the first one. Then a simple Wiener filter is used for deblurring.As the final part of this research, the real time video transmission over LAN is investigated. Since a ship is a relatively large space and it is required to setup many video cameras in all of the shipboard compartments linked with a unique monitoring station on the bridge, the local area network (LAN) connection of video cameras is a very important stage for the digital video-based fire detection systems.The real time video is captured from the camera of a remote computer and then the video stream is transmitted to another computer over LAN using TCP/IP protocol. The received video images are sampled and analyzed in the second computer and then the existence of fire is determined by using the fire character pixel detection algorithms. All fire detection experiments are carried out under the four different surrounding light conditions:bright, normal, dim and dark with and without fire.Frame rate of 15 frames per second (fps) is used for the transmission of video stream and therefore there are about 27 Mb/s of data flow rate which is within the available LAN Ethernet speed ranges. According to the results, no obvious network errors are found.
|
PDF Full Text Request