摘要
根据火焰独特的纹理特征,提出以块的LBP直方图特征为主并结合其他动、静态特征的火焰识别算法。首先用帧差法和RGB颜色高斯模型进行运动颜色检测,得到疑似火焰区域;再提取其红色分量统计特征、小波高频能量和LBP直方图特征;最后将特征向量输入SVM分类器进行火焰识别。实验证明,该算法误报率低、鲁棒性强,同时具有实时性,火焰视频检测率可达到96.2%。
A cco rd in g to uniq ue texture feature o f fla m e , we propose a flam e recognition algo rithm w h ich is m a in ly based on LB P histogramfeature and com bines other dynam ic and static features. F irst we use fram e difference and RG B colour Gaussian m odel to carry out m otion andcolour detection to get the suspected flam e area. Then we extract its red com ponent sta tistica l fe a tu re , high -fre que ncy wavelet energy and LB Phistogram feature. F in a lly we in p u t the feature vectors in to support vector m achine ( S V M ) cla ssifie r fo r flam e recognition. E xperim ents showthat th is algo rithm has low false ra te , good robustness and re a l-tim e p ro p e rty, the flam e video detection rate reaches 96. 2 % .
作者
张霞
黄继风
Zhang Xia;Huang Jifeng(College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 200234, China)
出处
《计算机应用与软件》
CSCD
2016年第8期216-220,共5页
Computer Applications and Software
基金
上海市教委科研创新重点项目(14ZZ125)
关键词
火焰检测
颜色高斯模型
LBP
小波高频能量
SVM
Flam e detection
Colour Gaussian model
LB P
H ig h -frequency wavelet energy
SVM
