摘要
针对传统火灾检测存在的对周围环境要求比较高,且研究对象多数属于室内或者商业设施等现象,提出了基于滑动窗口图像特征提取的多特征融合和SVM相结合的秸秆焚烧火灾检测算法。首先在YCbCr空间模型下使用Otsu(大津算法)对火焰图像进行前景检测,再对所检测到的前景使用颜色判别方法,得到候选火焰区域,然后使用滑动窗口在这些区域上进行移动,在每一个窗口内提取HOG特征、灰度共生矩阵特征、颜色矩特征,将这些特征分别送入SVM训练得到不同的分类器进行秸秆焚烧事件检测。最后根据投票方法将三种特征进行融合,最终检测出是否发生火灾。实验结果表明,该算法实现简单,识别率高,可达到86.67%。且由于算法基于火灾的静态特征,更能体现火焰的固有图像特征,与其他类型的火焰检测相比,适用性更强。
The traditional fire detection system has higher requirements of the surrounding environment, and research objects belongs to in- door or commercial facilities mostly. In order to solve the problem, a fire detection algorithm based on multi-feature fusion of image feature extraction of sliding window and the SVM classifier is proposed. First, in the YCbCr color space model, Otsu is used to detect the foreground from the flame image and the foreground is inspected by the color feature to obtain the fire candidate regions. Then three kinds of flame features including HOG,gray level co-occurrence matrix and color moment are extracted by using the sliding widow moved on the fire candidate regions,which are put into the SVM classifier to classify the window whether there is a fire. The final result based on the method of voting is gained. The experimental results show that the algorithm is simple and has higher recognition rate ,more than 86. 67%. Because it is based on the static features of the fire ,it shows inherent image characteristics of fire. Compared with other fire detection algorithms, it has wider application range.
出处
《计算机技术与发展》
2017年第5期69-72,76,共5页
Computer Technology and Development
基金
江苏省重点研发计划项目(BE2015707)
江苏省水利科技项目资金资助项目(2012034)
关键词
颜色模型
滑动窗口
单一特征提取
多特征融合
color model
sliding window
single feature extraction
multi-feature fusion
作者简介
聂建豪（1990-），男，硕士，研究方向为模式识别与图像处理；
李士进，博士，教授，研究方向为模式识别与数据挖掘。
