基于纹理特征和轮廓光流矢量的烟雾识别

Smog detection based on texture features and optical flow vector of contour

为了提高火灾探测的准确率和快速性,提出了基于纹理特征和轮廓光流矢量的烟雾识别算法。一方面为了获得更全面的纹理特征,建立图像金字塔,使用局部二值模式(LBP)和基于方差的局部二值模式(LBPV)结合的新方法分别提取金字塔不同层的纹理特征。另一方面是动态纹理特征,由于烟雾运动的湍流特性导致方向具有特定的一致性,改进了对全部可疑区域进行分析的方法,仅对可疑区域轮廓进行光流矢量分析,降低运算量。将静态纹理特征和动态纹理特征输入支持向量机(SVM)中进行识别。采用"静—静—动"的新型识别方法,实验结果表明:该算法能够及时准确报警,可靠率高。

In order to improve accuracy and rapidity of fire detection, a smog detection algorithm based on texture features and optical flow vector of contour is proposed. On one hand, to obtain more comprehensive texture feature, image pyramid is constructed, a new method combines local binary pattern （LBP） with LBP based on variance （LBPV） are used to extract texture feature of different level of the image pyramid. On the other hand, due to special turbulence characteristics of smoke movement ,its moving directions are with a certain consistency ,improve the method of analysis on all suspicious area, optical flow vector analysis is carried out only on suspicious area contour,which greatly reduces computational complexity. These features are input to SVM for recognition. ＇ Still- Still-Moving＇ approach is used, experimental results prove that it can timely and accurately gives alarm and it has high reliability.