The Video Type Fire Detection Technology Based On Fpga And Method Of Research

Video fire detection system, based on machine vision, is able to be applied toalmost any place. Most video fire detection systems are designed with microprocessors.However, serial instruction and serial data processing of microprocessors make arestraint to the operating speed. On the contrary, FPGA is capable of parallel dataprocessing, making an advantage in digital image processing. Even if a large amount ofparallel image data can be processed in real-time by FPGA.This paper is aimed to implement a video fire detection system capable ofacquiring image data, preprocessing image and recognizing flame in real-time, with acore of FPGA. An infrared filter is set in front of the camera, in order to separate flamefrom the low-temperature background, and image data from sensor need binaryzationand morphology processing before recognition. The mathematical models of areafeature, circularity feature and corner feature of flame image are set up in this paper andverified by experiments. The result of experiments shows that all the three features offlame image are unconditional. Bayes decision-making theory is used to analyze how todecrease the misjudgement rate in flame recognition, and the analysis indicates thatincreasing the features will be a most direct and effective path to lower themisjudgement rate. As a result, a flame recognition algorithm of multiple features isproposed. An experiment shows that misjudgement rate of the proposed algorithm ismuch lower than that of single-feature algorithm. At the end of the paper, the method ofimage preprocessing and the flame recognition algorithm are translated into VerilogHDL. The key of translation is to parallelize the algorithm. Pipeline and LUT operationsare taken to reach the top running speed of FPGA. Simulation wave and test resultshown in this paper prove that the implementation of video fire detection on FPGA isfeasible.