基于块分割和SVM的视频火灾烟雾识别

Fire-smoke Recognition Based on Block Segmentation and Support Vector Machine

研究火灾早期预警问题,为了解决现有图像型火灾烟雾检测算法对环境适应能力不强和探测准确性不高等问题,采用分块和背景差分,分析烟雾在RGB和HSV颜色空间色彩和亮度的变化规律获取疑似烟雾区域;以运动主方向和疑似烟雾与背景对应区域的高低频能量比作为火灾烟雾识别依据,构建SVM分类器,最终实现烟雾判决。并与BP神经网络模型火灾烟雾探测方法进行了比较分析,结果表明,改进的算法具有较强自适应性,识别率高,实时性强,鲁棒性高,可适用于多种火灾探测场景。

Dubious fire smoke regions were obtained by block segmentation and background difference, and color and lighteness changes of RGB and HSV spatial model were analysed for resolving the problems of fire smoke detec-tion, such as weak entironment daptability and low veracity. Primary orientation angle and energy ratios of highe/low frequency of dubious smoke regions were acquired as the input vectors of SVM classifier in order to realize smoke rec- ognition. The experimental results show the method has strong adaptability and higher accuracy, and can be applied in a variety of fire detection scenarios.