Research And Design Of Fire Detection System Based On DSP

Fire is one of the common disasters that threaten the safety of people's life and property,so it is very important to study the technology of fire real-time detection and early warning.The traditional sense based on temperature, smoke and other fire detection methods can not meet the application requirements in large space, multi interference and other occasions. Information collection of image type fire detection technology has no direct contact with the outside world and the wide range of characteristics. Because of this information collection features,the image based fire detection technology has become a large space, multi-jamming site fire detection method. The image prcossing system based on computer has the advantages of high cost and difficult to install, so the embedded system is the best choice for image based fire detection system. The purpose of this paper is to set of iamge fire detection system based on DSP to realize the fire detection in the target area.Firstly, based on the analysis of the research at domestic and abroad and combined with the embedded technology, image processing and pattern recognition, this paper has designed the system of fire detection modularity. It includes image processing module, image acquisition module, system memory module, etc. Among them, the image acquisition module includes the design of image capturing apparatus and image collection.The system memory module includes the design of SDRAM, FLASH and CPLD's function.This paper has improved the traditional method of image preprocessing of fire. In this paper,the preprocessing procedure is as follows. Firstly,we apply the iterative method to binaryze the image of fire. Secondly,the image is labeled with connected domain. At last,we apply a new filtering method to process the image.Based on the completion of the preprocessing of image of flame and combined with the characteristics of the original image,several criteria for early fire were proposed in this paper,including area change percentage, relative stability, circularity and the number of sharp corners of fire. After that, we extracted six values of features on the platform. Combined with the identification requirements of system, this paper designed BP neural networks with three layer, and we have done a lot of experimens to train the networks for obtaining weights and biases of neural. After the training of neural network, we moved it to the platform of DSP to recognize the fire.The hardware and software of the system were realized by the method of fire detection based on imge and the identificaotion method of fire. A lot of experiments were carried out to verify the feasibility of the system. The results showed that the system designed by this paper can detective fire accurately and rapidly in large space and multi disturbance palce.At last, we dicussed contributiuons of our study and the designing of the system, and point out the developing trendency.