Fire Features Detection System Based On The Video Monitoring

Fire is a kind of disaster which is common and multiple, and it threats human being and properties greatly. Doing research work on how fire happen and develop, specifically on how to detect and prevent fire, is very important to protect human being and properties. At present, primary means of fire detection is based on sensors, such as smoke detectors, sensitive fire detector, heat detector, and so on. But detectors based on sensors have many drawbacks, for example, its monitoring range is small and it is easily failure, and so on. Based on the above considerations, I established my goals, which is fire features detection based on the video monitoring.At present, at home and abroad for the characteristics of video-based monitoring of fire detection studies have been few, but the specific monitoring technology has not really applied to our production and life, this is because there is not a fruitful research can achieve our requirements - and effectively carry out real-time fire monitoring and control, so this route has yet to explore all of us.1. The extraction of the fire suspected areaResearch fire monitoring, it is first images of fire analysis and research of the combustion flame. Flame is a dynamic process, reflected in the image is a moving objects, analyze the image foreground and background, as the flame test basis. The extraction of foreground and background is important, only the extraction of effective flame area suspected to be able to identify the characteristics of the flame to the next step. The prospects for extraction of common methods are: the frame difference method, reducing the background law, but these two methods have some limitations and can not achieve our desired results; it does not apply to the system.Kwang-Ho. Cheong et al[12] put forward the prospect of a continually updated background extraction algorithm, this paper to improve on this method, in the continually updated based on the background, through the introduction of the learning rateλto reflect the changes of the background image of the scene The response has been very good results.2. Analysis of fire features1)Flame color features: first, analysis of video images of the flame through the observation of a large number of video images of flame and non-flame video images, was found to contain the fire video image does have a certain pattern: the flame image pixel the value of red channel > the value of green channel > the value of blue channel, and the red channel value is higher; In addition, a large number of experiments show that, in the HSV color space, through the S value analysis can also distinguish the flame pixel points.2) The flame flicker frequency characteristics: the flame burning so that we often feel the flames of violent beating, the beating frequency is called the flame flicker frequency. In particular, the edge of the flame part of the video in a few seconds will often appear to a non-fire by the fire or by a non-fire change back and forth to the fire.3) Flame image correlation features: correlation detection algorithm is based on the study of dynamic characteristics of the flame detection algorithm was first proposed by Ollero[17] proposed the use of two adjacent frames the correlation between the detection of forest fires, but this method is not conducive to the practical application of the mutation will produce relatively large number of false negatives and false alarm. Later, Shen Shi-Lin[18] on the correlation detection algorithm to be improved and will be a period of time of each frame and continuously updated background image to do correlation analysis to study the correlation coefficient changes over time resulting from fluctuations in the orderly conduct of fire, flame detection, revenue to the more obvious detection.4) Flame image wavelet transform features: wavelet analysis, we can analyze the characteristics of various changes in the original signal, and further used for data compression, noise removal, feature selection and other fields. In this system, the image is two-dimensional signal, the wavelet transform is equivalent to the second one-dimensional wavelet transform signals, using two-dimensional Mallat algorithm[20] The decomposition process can be expressed as: the first one-dimensional wavelet transform is equivalent to the signal rows images transform; the second one-dimensional wavelet transform is equivalent to the image signal of the column transformation, after the completion of decomposition produces four sub bands LL, LH, HL, and HH, respectively, corresponding to low-pass filtering the signal, the horizontal high-pass filtering signals, Vertical high-pass filtering the signal and diagonal high-pass filtering the signal.5) Flame image cusp features: through a large number of study and observation to find out the flame with the angular features of such a shape [21], performance in this system is based on the shape of the detection; the main outlook is to measure the number of marginal cusp. The current system cusp detection algorithm assumes that such a pattern consistent with a cusp: the existence of a vertex or very short Vertex side; there is greater than a certain length of the two corner edges; corners to meet a range of side angle.3. Analysis of fire features under infrared cameraInfrared cameras are used for flame detection, under the five characteristics of the flame detection, namely: the flame flicker frequency, the flame wavelet analysis, angular features of the flame, the flame DCT transform analysis, and color characteristics of the flame, which flame flicker frequency, Flame of wavelet analysis and flame cusp characteristics and the flame under normal camera detected the same, there will not repeat here, analyzes the prospects for infrared camera under the extraction and the other two characteristics.1) The prospect of extraction under infrared camera: Because infrared brightness of the flame characteristics of the video is very clear that to meet the brightness set by extracting the pixels, we can extract most of the suspected fire area, so the prospect of flame detecting infrared video extract is based on the brightness Extraction of flame prospect. Infrared video images, the brightness of the flame region to become very significant feature, extract video screen high-brightness area can ensure that the extraction of the flame region.2) Flame color features: Since the system uses the infrared camera instrument has automatically adjust the brightness depending on the video frames the numbers of infrared lamps feature, when the video frame there is a large bright area, this region near the convergence of ordinary color camera to capture images some properties that will exist some of the non-gray pixels. The observation that the dark, the brightness of the flame combustion infrared camera were able to shut down part of the infrared light, resulting in partial red flame appears around the region pixels. The multi-segment video test, these reddish pixel differences between the maximum and minimum values larger. According to this feature, the system is designed infrared video of the flame color detection algorithm.3) Flame image DCT Transform Features: Flame video, the prospects for the height of the image is constantly changing, and take advantage of this law can also be right to judge the flame. ZHANG Jin-Hua[25], etc. On this basis, put forward a novel and efficient solution: instead of using the flame height of the flame flicker frequency. The characteristics of flame flicker frequency there is a certain connection with the general disturbance and there is a big difference. Records of the flame height of a period of time to conduct discrete cosine (DCT) transform[26], one of the characteristics identified as the flame.In short, this article is based on the basis of previous studies, carried out targeted improvements to realize a video-based monitoring of fire detection system features. The system has a number of advantages: real-time, able to issue a warning at the first time when fire happen; the accuracy of detection algorithm for improved effectively improve the detection rate and reduce the false detection rate; system is stability, timely release of useless memory, clean up the memory space; portability, the system can be ported to run on a variety of platforms; anti-interference, compared to a variety of sensors-based monitors, cameras to monitor the use of more anti-interference and the monitoring of large areas.