Study And Realization Of Multi-Feature Based Fire Image Detection

The using of fire is an important symbol of human towardsciviliazarion, however, fires out of control may result in a huge loss of life and property. Therefore, the fire detection technique is becoming the focus of attention, and its development is important significance to the progress of human and the stability of the society.Currently, temperature detection and smoke detection are the widest and the most mature technology, but such detection technology cannot satisfy the fire detection requirement of the places like large volume buildings, tunnel constructions, and complex buildings.In recent years, with the development of the computer technology and information processing technology, video-based detection technology gets rapid development it has the advantages of fast response, wide detection range, low environment pollution etc. Comparison with traditional fire detection systems, it has broader prospect. At present, some products with infrared or ultraviolet radiation detection technology have appeared on the market. On one hand, this kind of optical detectors is expensive because of the optical equipment; on the other hand, the false alarm and failed alarm are difficult to control and to promot. The existing fire video detection technology is difficult to remove the interference in complex sceneseffectively. This paper provided insight on the flame of visible light spectrum for video detection technology, improved the method of extraction on moving target, target feature extraction and target recognition, proposed a method of accurately obtain the flame color and target tracking technology, also developed a flame image detection system. This main research highlights in this paper are in the following aspects:1. Although the CCD camera quality is better than the CMOS camera in image collection, this paper give up the traditional fire detection with CCD camera, and chose the application more extensive cheap CMOS camera, based on MATLAB, to realize the high-speed video collection of the monitoring scenes.2. Put forward a kind of method foraccurate extraction of the flame, to refine n-heptane and ethanol fire in different light environment for statistical analysis, obtained n-heptane and ethanol color model, then on this basis, combining previous research method, conducted a lot of experiment and statistics, finally obtained an effective color decision model.3. Extensive researched on typical algorithms of moving objects detection, and analyzed their advantages and disadvantages. The paper introduced independent component analysis (ICA) technology into the fire detection field, and introduced the basic model of the ICA, FastICA model, GICA algorithm in detail. With the method of the cumulative difference image, and put forward the "does not exist background modeling" accumulation fast GICA algorithm. Additionally, the paper Constructed a data storage structure for multi-target extraction, tracking and recognition, and proposed a kind of the multi-target tracking algorithm based on movement region and data delay technology, finally, realize multi-target tracking and its corresponding characteristic value and the transfer of the classification in fire scenario.4. Research the existing results of video flame detection technology, and the typical flame image method of feature extraction, we introducing Statistical Landscape Feature into fire video detection, on its basis, we put forward an improved faster computing method of Statistical Landscape Feature, and realized the flame feature extraction. In addition also built up flame multi-characteristics BP neural network classification model, realize the flame recognition.5. The n-heptane and gasoline fire experiments were designed, and acoustic measuring instrument was introduced for burning sound measurement, then, the rule of flame voice frequency is given out, and at the same time, based on the existing flame video frequency rule, the paper discusses some basic problems between the video frequency and the voice frequency. This paper proposed a fire detection system based on image and sound detection technology, also designed the Flame Image Detection System, then real-time monitoring of scene and multi-scene video for test algorithm performance, test indexes include the response time and fire detection rate.