基于图像分离的视频烟雾检测方法

Smoke Detection Method in Video Based on Image Separation

为提高视频中烟雾检测的准确率,提出一种基于图像分离的检测方法。通过高斯混和模型和统计非参数方法建模背景、消除阴影和提取前景区域,在得到前景的基础上结合暗通道得到候选烟雾区域,对候选区域进行分块处理,利用大气散射模型提取该区域内每块图像帧中的烟雾成份,并逐一提取局部二值模式特征,使用支持向量机进行分类和判别,同时采用亮通道进一步加快烟雾成份分离速度。实验结果表明,该方法能够提高烟雾检测准确率和检测速度。

For improving the accuracy of detecting smoke in video, this paper presents a method using image separation. Firstly, it uses Gaussian Mixture Model （GMM）and Statistical Non-parametric approach （SNP）to model background, eliminates shadow and extracts foreground image. Meanwhile, it combines dark channel image to get candidate smoke regions. Secondly, it extracts smoke component from candidate regions. Thirdly, it extracts Local Binary Pattern（LBP） feature. Finally, it uses Support Vector Machine （SVM） to classify and identify. To speed up the smoke component separation ,this paper uses bright channel. Experimental results show that the proposed method can improve the speed and accuracy of smoke detection.