针对公路隧道火灾的早期探测问题,提出了一种基于多特征融合的视频火灾火焰识别方法。利用帧间差分法提取运动目标,根据RGB和HSI空间的火焰颜色统计模型分割疑似区域,用疑似区域的圆形度、矩形度、尖角数、尖角数的变化率和面积的变化率特征值构成特征向量,作为支持向量机的输入向量,采用人工蜂群算法对支持向量机的模型参数进行优化选取。利用公用视频和拍摄视频对训练得到的支持向量机分类器进行测试,试验结果表明,该方法能够实现在隧道场景及多种干扰条件下的早期火灾火焰识别。
Aiming at the early fire detection in road tunnel, a video fire flame recognition method based on multi-feature fusion of video flame is proposed in this paper. The frames difference algorithm is used to extract the moving target, the suspicious region is segmented by the color statistical model of flame in RGB and HSI spaces. The character vector is formed using the eigenvalues of circularity, rectangularity, number and its variation rate of sharp angle, area variation rate in the supicious region, which is used as the input vector of support vector machine (SVM). The artificial bee colony (ABC) algorithm is used to optimize model parameters of the SVM. The classifier of the SVM is tested by means of public videos and taken videos. The experimental result reveals that the proposed method can effectively recognize the early fire flame in road tunnel with different disturbances.