摘要
视频中类似烟雾的区域很大程度上增加了烟雾检测的误差,为了提高烟雾检测的准确性,利用BP神经网络分类器,一种基于小波特征、烟雾纹理特征以及Y分量均值特征相结合的烟雾检测方法被提出。首先对视频序列进行运动区域提取;然后对疑似区域提取小波特征、纹理特征以及Y分量均值特征,形成一种新的多特征组合向量;最后将特征向量输入到BP神经网络分类器进行检测。实验表明,通过这种组合特征的方法,检测结果更有效。
Smoke-like regions greatly increase smoke detection errors in the video. In order to improve the accuracy of smoke detection, a smoke detection method based on BP neural network is proposed, combining the wavelet feature, smoke texture feature and mean of Y component pixel value. Firstly, moving regions in the video sequences are ex- tracted;secondly, the wavelet feature and texture feature of suspected regions are extracted, then a new kind of multi-feature vector is formed. Finally, feature vector is input into the BP neural network classifier for smoke detection. The experiments show that smoke detection results are more effective by combing various features.
出处
《计算机科学与应用》
2013年第5期239-243,共5页
Computer Science and Application
基金
中央高校基本科研业务费专项资金资助。