基于多特征融合运行期均值法的烟雾检测算法

Fire smoke detection method based on multi-feature running average method

烟雾检测能够有效预防火灾的发生,以此为出发点,解决了烟雾运动区域出现的"空洞"现象,从而实现了快速、准确地检测视频中烟雾的区域。利用烟雾的非刚体特点,融合颜色、背景模糊以及运动方向的特征对背景更新过程进行建模,然后结合连通域面积消除小干扰区域的影响,提高了基于单一特征的运行期均值法的鲁棒性。实验结果表明,改进的方法能够准确、完整地检测出视频中的烟雾区域,并且处理速度也有所提高。

For detecting smoke which is sign of fire, the ＂hole＂ phenomenon which is in the task of smoke detection is resolved in this paper. Considering the non-rigid nature of smoke, the color, blurred background and motion direction are all included to mod- el the process of renewing background. And then the small noise regions are eliminated based on connected area. The improved run-average method is more robust than the one based on single feature. The experimental results show that the proposed method can accurately and efficiently detect the smoke area in video.