【作者】 张瑞；

【导师】 吴冬梅；

【作者基本信息】 西安科技大学 ， 信号与信息处理， 2015， 硕士

【摘要】 火的出现,使人类区别于其他物种的发展,开辟了文明历史的新纪元。如果能够准确使用火会造福人类,一旦放松对其的警惕,导致无法被控制,则会带来很严重的损失。基于硬件的传统火焰检测器,只有达到要求时才对火灾做出响应,不能够做到及时有效的报警。因此,人们把眼光聚集到了实用性强、响应快的智能视频系统上。本文主要研究了基于火焰动态模型融合的火焰检测算法,对提取的前景目标结合RGB和HSI颜色模型确定出视频图像中的疑似火焰区域。然后结合火焰的动态特征对疑似火焰区域提取面积变化率、规则度和重叠度三个特征值,分别设定阈值进行分析判断。然后将三个特征值放入火焰动态模型中进行融合,设定权值对疑似火焰区域进行综合分析判断。最后,标示火焰区域。实验结果表明,算法能够达到实时性,对环境有一定的适应性,对类似火焰物体、发光源有一定的抗干扰能力。当火焰初期火焰动态特征不显著时检测效果明显,当火焰特征显著时,检测结果比较准确。相比利用帧间差分提取运动目标结合RGB颜色空间进行判别的方法,该算法漏检率明显下降,基本不存在误检现象。更多还原

【Abstract】 The discovery of fire, so that human development is different from other species, has opened up a new era in the history of civilization. If it used correctly will benefit mankind, once relax their vigilance, resulting in not being controlled, it will bring very serious losses.Traditional hardware-based flame detector are responding fire only to meet the requirements, can not be done on a timely and effective fire alarm. Therefore, people gathered to look into the practical, fast response intelligent video systems.This paper studies the flame detection algorithm based on a fusion dynamic model of flame, combined with the prospect of extracting the target RGB and HSI color model to determine the video image of the suspected fire area.Then combine the three eigenvalues dynamic characteristics of the suspected flame region extraction rate of the area change, roundness, and the similarity,setting the threshold to analyzed and determine respectively. Then put the three eigenvalues of the dynamic model into flame fusion set weights suspected comprehensive analysis to determine the flame region. Finally, marking the flame zone.Experimental results show that the algorithm can achieve real-time, there is a certain adaptability to the environment, to similar objects flame or emitting source has certain anti-jamming capability. When the dynamic characteristics of the initial flame is detected no significant effect is obvious,the test results of the flame characteristics significant is more accurate. Compared to the method of using frame difference to extract moving objects combined RGB color space identification, the algorithm undetected rate decreased, the phenomenon of false detection basically does not exist.更多还原