基于新的运动特征的火焰检测方法

Fire Detection Method Based on New Moving Feature

当前火焰检测方法在刻画火焰特征时忽略了火焰的运动方向信息,从而对于复杂背景环境下的火焰目标可能会产生错误的检测结果。为了更准确的刻画火焰,在层次火焰检测方法的基础上,对多个层面,融入光流法获取的火焰方向信息,提出一种新的火焰运动特征,从而实现火焰检测。首先利用光流获得火焰候选区域每个点在四个方向区域的分布,统计不同区域像素的比例信息获得火焰方向特征;然后在时空层次上结合方向特征进行分析形成火焰频率特征;最后把方向和频率特征结合形成火焰的运动特征,利用核支持向量机(Kernel-SVM)对该特征进行训练,得到火焰检测模型。实验结果表明,火焰运动特征能显著提高火焰检测的准确性和降低误报率。

In order to more accurately characterize the fire feature and improve the fire detection in complex scenarios, a novel moving feature is proposed with the information of optical flow based on the multi-level hierarchical fire detection. First, the direction of each pixel in the fire candidate area is obtained by computing the optical flow and the percentage of pixels whose moving directions fall into four discrete parts is calculated as the orientation fea- ture. Then, the frequency information of change in a period of time is obtained with the information about the orienta- tion of the fire as the frequency feature based on the hierarchy of temporal-spatial. In the end, a new fire moving fea- ture is obtained based on the frequency and the orientation feature. By using support vector machine （kernel-SVM） classifier with the input of the moving feature, the fire detection model is obtained. The experimental results confirm that the feature can significantly improve the accuracy of fire detection and impressively decrease the false alarm rate.