基于水平集的火焰分割  

Level Set Based Flame Segmentation

作　　者：江文婷[1] 龚小谨[1] 刘济林[1] 

机构地区：[1]浙江大学信息与电子工程系,浙江杭州310027

出　　处：《杭州电子科技大学学报（自然科学版）》2013年第4期27-30,共4页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：国家自然科学基金资助项目(61001171)

摘　　要：基于视觉的自主火灾检测在火灾预警方面有很重要的应用价值。该文提出了一种基于水平集的火灾检测方法,通过准确分割图像中的火焰部分来识别火灾。该方法将火灾方面的特定信息集成到水平集框架中:首先得到火灾的概率观测模型,然后通过考虑最大后验概率和边缘线索形成水平集的能量函数,最后通过最小化能量函数,使用水平集的轮廓能准确地划清火焰的区域。算法在一个含有各种火灾场景的综合性图像集上进行了测试,实验结果验证了算法的有效性。Although level set methods are widely used in object detection and segmentation,they are seldom applied to fine detection in literatures.In this paper,it presents a level set based method for fire detection,especially,for precise flame segmentation which is essential in fire metrology.The proposed method incorporates fire specific knowledge into the level set framework.In particular,it first learns probabilistic observation models of fire and background from training data.Then the level set energy functional is formulated by considering both the maximization of the energy functional,the obtained level set contours can precisely delimit flame regions.The proposed algorithm is tested on a comprehensive image set containing diverse fire scenarios.The experimental results validate the effectiveness of our algorithm.

关 键 词：火焰分割 火灾检测 水平集 概率框架 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]