Research On Real - Time Monitoring Method Of Smoke Based On Video

In the field of fire safety, in order to make the interests of the people far away from the fire threat, there have been a lot of fire detection means. At present, the market mainstream fire detection products are based on sensor type, in a variety of private houses, large shopping malls, factories and other places in the workshop, the distribution of these sensors can be seen everywhere. The mainly working principle is generated according to the fire when it happens with the smoke and flame, use the corresponding variety of light, temperature, and chemical sensors. To detect the abnormal and abnormal report will be given to the alarm device which connected to sensor and then deal with fire equipment. The problem of these sensor type fire detection technology mainly exist in the three aspects: one is a single detection data, the sensor for the detection of fire data usually just contains a little smoke or flame information; the second is its small monitoring range, sensor can only detect a small independent space, in a large space the device has strong limitations; the third is the lack of stability, sensor devices prone to old status and is easy to be ambient interference occurred missed detection and false alarm.Taking into account the defects of the above sensors in the fire detection, the current researchers began a lot of research in the visual direction of fire detection. Mainly using the pattern recognition and image processing technology, the visual features of fire extraction, analysis and study, in terms of smoke commonly used features such as color, motion, frequency, texture, etc.. In this paper, based on the reference and study of a large number of previous hard work, in view of the smoke, put forward two kinds of new smoke video detection algorithm.A smoke detection algorithm based on multi feature fusion and SVM classifier, in which the video is divided into blocks, so as to facilitate the extraction of smoke from an image block. The algorithm based on the texture, color, fluctuation characteristics of the smoke, put forward the vector of NR-ULBP and IQ-OLBP texture feature; according to the smoke color distribution of smoke Col color feature vector; smoke Wav feature vector is proposed on the basis of the volatility of the smoke, finally will be the four feature fusion, a characteristic Feature vector. Feature vector not only contains a lot of the smoke information, but also compared with similar features, Feature vector has shorter length, stronger decision. The algorithm also according to smoke’s spatial and temporal characteristics, then proposes a method to eliminate the interference. Experimental results prove the classifier algorithm can ensure the real-time video smoke detection and keep the algorithm accuracy and stability.The other is based on motion and color video smoke detection algorithm, the algorithm also uses the block, but here is to ensure that the continuity of the movement area. Firstly based on the adaptive background updating model, this algorithm proposed a background updating model based on block, the model is not just divided the image into blocks, also according to the difference between foreground and background on the block, we put forward a method for adaptive background adjusting and updating. Then, the color characteristics of smoke in YIQ space is adopted, a smoke color normalization model is put forward, and the color of the pixels is verified. At last, using the smoke discrete motion model, the candidate blocks are judged to determine whether they are in accordance with the smoke movement. Experiments show that the method is simple, practical and accurate, and can be used in some practical applications.