基于HOFHOG和RDF的火灾区域探测

Detecting fire region based on random decision forest and HOFHOG features

为探索基于时空特征的室内外火焰和烟雾的识别,分析在时空块内不同通道下的光流直方图,探索火灾区域的梯度方向直方图的静动态特征描述方法,采用各分量相邻帧差组成时空块间特征的统计数据,反映各特征的空间和时频分布属性,对随机决策森林树训练过程中的参数、性能进行选择和分析,探测火焰与烟雾区域各特征的空间分布和时序关系,由决策森林投票给出更逻辑合理的判断。实验结果表明,该方法在火灾探测系统中表现出稳定的识别精度。

To probe the recognition of flame and smog based on colors and spatial-temporal features in indoor and outdoor illumi-nation,an analysis way for histograms of oriented optical flow in different channels was proposed.How to describe fire region using static and dynamic features of histograms of oriented gradients was researched.A way of adopting block?s frame difference statistic attributes as spatial-temporal fusion feature was presented.Parameter selection and performance analysis for random de-cision forest classifier training based on feature subsets were presented via relief feature selection.Flame and smog region were detected simultaneously and a more logical alarm judgment was submitted by decision tree forest voting according to detected re-gion?s spatial-temporal distribution and relations.Experimental results show that the flame and smog detection system based on random decision forest classifier has more stable and higher accuracy.