【摘要】 提出一种基于烟雾多特征融合的图像型早期火灾烟雾检测方法.利用混合高斯模型背景差法提取前景中的疑似区域,根据早期烟雾在RGB和HIS空间的颜色模型提取疑似区域的颜色特征值,通过二维离散小波变换提取背景模糊特征值,通过计算疑似区域像素数与其最小外接矩形面积比提取轮廓不规则特征值.根据文中提出的综合判据,融合三种特征值进行烟雾判别.对公用视频和拍摄视频进行测试.结果表明,该方法能够适应室外场景变化及多种干扰条件,有效检测早期火灾烟雾.更多还原

【Abstract】 An early fire smoke detection method combined of color,background blurring and contour disorder features was proposed. A mixed Gaussian model of the background was established in RGB space,while the suspicious region was extracted by comparing the current frame with its background reference model. According to color model of early smoke in RGB and HIS space,the color feature of suspicious region was obtained. The background blurring feature was obtained by two-dimension discrete wavelet transformation,the disorder feature was computed by calculating ratio of pixels in suspicious region to the area of minimum bounding rectangle. The features extracted above were used to determine whether the suspicious region was smoky according to a joint judging rule. Experimental results on public test videos show that the proposed method can improve the accuracy of smoke detection and can be used in outdoor environment.更多还原