基于混合高斯模型与小波变换的火灾烟雾探测

Smoke detection method based on mixed Gaussian model and wavelet transformation

本文提出了一种基于混合高斯模型和二维离散小波变换的图像型火灾烟雾探测方法。在RGB空间使用混合高斯模型对背景进行建模,通过当前图像与背景参考模型的比较提取可疑区域,分割可疑图元。对可疑图元进行特征提取:在灰度空间通过二维离散小波变换提取高频和低频能量特征值;在HSV空间提取颜色饱和度特征值;利用图像帧序列提取边界闪烁频率特征值。综合各特征值进行烟雾判断。使用烟雾视频图像进行烟雾判断实验,结果证明了此种方法的有效性。

This paper presents a new fire smoke detection method based on mixed Gaussian model and wavelet transformation. A mixed Gaussian model of the background is established in RGB Space. And the suspicious region is ex- tracted by comparing the current frame with this reference model. Then the feature of the suspicious region is investigated. The high/low frequency energy eigenvalue is obtained through discrete wavelet transformation （DWT） in grey space, the color saturation eigenvalue is obtained in HSV space and the boundary flicker frequency eigenvalue is obtained using the frame sequence. The eigenvalues are used to determine whether the suspicious region is smoke according to judge rules. Experiments were carried out using real images of the fire smoke, and the result proved the validity of the method.