摘要
烟雾检测对火灾早期防范非常重要,传统的火灾探测技术主要利用传感器对火焰和温度进行识别,其每一个传感点只能检测到布控点周围的局部空间,对于开放空间等特殊场合难以发挥作用。为了克服传统火灾检测存在的误报率高等缺点,文中提出一种基于烟雾多特征融合技术的图像型火灾检测方法。该方法首先利用背景减除法获取普通CCD摄像机拍摄的疑似火灾烟雾区域,然后再从时域和频域着手,提取火灾烟雾的轮廓不规则特征、背景模糊度特征和纹理特征作为神经网络的输入信号,同时采用sigmoid函数将输出归一化,最后通过对BP神经网络训练完成火灾烟雾的多特征融合,并对来自网络的火灾视频进行测试。实验结果表明:图像型火灾检测方法能够准确快速地识别火灾烟雾,达到早期预警的目的。
The smoke detection is very important for the prevention of early fire,the traditional fire detection is a technology that uses a sensor to identify the flame and temperature,each sensor can only detect dispatched around the local space,for the open space and other special occasions,difficult to play a role. In order to overcome the defects of traditional fire detection has disadvantage of high false alarm rate,a fusion technology of image fire detection method based on multi feature of smoke was proposed. The method uses background subtraction method to obtain the ordinary CCD camera shooting suspected fire smoke regions at first. Then from time domain and frequency domain,the fire smoke irregular contour feature,background extraction fuzzy features and texture features are extracted as the input signals of neural network,also with the sigmoid function will output a normalized. Finally through the training of BP neural network,complete fire smoke multi feature fusion,and carry on the test of fire video network. The results showthat image based fire detection method can accurately and quickly identify the fire smoke,and achieve the purpose of early warning.
出处
《计算机技术与发展》
2016年第1期129-133,共5页
Computer Technology and Development
基金
黑龙江省自然科学基金项目(C201244)
关键词
烟雾检测
轮廓不规则特征
模糊度特征
纹理特征
特征融合
smoke detection
irregular contour feature
ambiguity characteristics
texture feature
feature fusion