Multi-feature Flame Detection Based On Video Image

The fire always causes casualties and property damage, therefore, the accurate detection of the fire is the focus of most concern. The appearance of fire detector marks a new era of fire detection, which has important implications for human progress and social stability.At present, in the little space and ordinary place, the conventional thermal detector and smoke detector dominate the dominance, and the technological level of it is already very mature. But for large space and the place with disturbance, the conventional detectors cannot meet the requirements of detection, so it is in urgent need of an effective detection method.With the development of image recognition and computer technology, the fire detection technology based on video image is born at the right moment. It has the characteristics of quick response, wide detection range, energy-saving environmental protection and so on, so compared with the conventional detector, it has a wider foreground. Now there are some molding products of similar products at home and abroad, but because of the expensive cost, it is hard to popularize. The types of existing video fire detection technologies are varied, but only with the single static or dynamic characteristics of the fire, it is difficult to detect the fire accurately. In this paper, the fire detecting system is established with the characteristics of static color, textural and dynamic area of growth, combining the flame recognition by BP neural network.The main work and research results in this dissertation are as follows:(1) There are various extraction methods of color feature. In this paper, it is established for the flame color in HIS color space. And get four frames images per second with the collected video, with the color feature extraction for signal frame figure by video capture, the area of suspected flame is obtained.(2) The boundary of suspected flame is obtained by segmentation of its area with edge detection technology. The number of white pixels on the binary image is equivalent to the area of the suspected flame. Then on matlab platform, the statistical tracing point is conducted for the image segmentation area size in 3 seconds. And the slope of the function is calculated by linear fitting to judge the size, if it is greater than zero, the area of suspected flame is considered increased, then is considered decrease or remain the same.(3) The characteristics of texture and color channel of the images are extracted, and the BP neural network is established. The 220 images, which are downloaded, filmed and video captured, are used to train network parameters, and using 80 images to test the network, then get a effective neural network.(4) The recognition technology of color features, flame features and BP neural network are summarized, the fire detection system of multi-feature for recognizing fire based on video image is proposed. The simulation studies are given by writing programs and simulated test on Matlab platform. The effects of the proposed system with the test of 11 videos is better.