一种分步的融合时空信息的背景建模  

A Stepwise Background Subtraction by Fusion Spatio-temporal Information

作　　者：储珺[1] 杨樊[1] 张桂梅[1] 汪凌峰[2] 

机构地区：[1]南昌航空大学计算机视觉研究所,南昌330063 [2]中国科学院自动化研究所,北京100190

出　　处：《自动化学报》2014年第4期731-743,共13页Acta Automatica Sinica

基　　金：国家重点基础研究发展计划(973计划)(2009CB320902);国家自然科学基金(61263046);中国航天科技集团公司航天科技创新基金(CASC201102)资助~~

摘　　要：自然场景中的光照突变和树枝、水面等不规则运动是背景建模的主要困难.针对该问题,提出一种分步的融合时域信息和空域信息的背景建模方法.在时域,采用具有光照不变性的颜色空间表征时域信息,并提出对噪声和光照突变具有较好适应性的码字聚类准则和自适应背景更新策略,构造了对噪声和光照突变具有较好适应性的时域信息背景模型.在空域,通过采样将测试序列图像分成两幅子图,而后利用时域模型检测其中一幅子图,并将检测结果作为另一幅子图的先验信息,同时采用马尔科夫随机场(Markov random field,MRF)对其加以约束,最终检测其状态.在多个测试视频序列上的实验结果表明,本文背景模型对于自然场景中的光照突变和不规则运动具有较好的适应性.In a natural scene, it is difficult to create a background model for the presence of illumination variation and irregular motions including waving trees, rippling water, etc. This paper proposes a new stepwise algorithm by fusing spatio-temporal information. In the time domain, we characterize the temporal information in the color space which is invariant to photometric changing. On this basis, we propose a clustering criterion of codeword which is adaptive to noise and illumination variation, and present a novel adaptive background updating strategy. Then a temporal information background model which has a better adaptability to noise and photometric invariants is constructed. In the spatial domain, we first divide the test frame into two sub-images by sampling and then utilize temporal information to detect one of them. Yhrthermore, we regard the detection results as priori information of the other sub-image and adopt Markov random field to restrict it simultaneously, then detect its state. Extensive experiments are conducted on several test video sequences. Compared with the mixture of Gaussians （MOG）, standard codebook model （SCBM）, and improved codebook model （ICBM）, the results show that out algorithm has better adaptability to the illumination variation and irregular movement in natural scenes.

关 键 词：时空背景模型 前景检测 马尔科夫随机场 码本 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]