基于空时置信关系的运动检测方法  

Motion detection based on spatio-temporal confidence relationship

作　　者：范志辉[1] 陈曦[1] 王辉[1] Fan Zhihui;Chen Xi;Wang Hui(Henan University of Science ＆ Technology,Luoyang Henan 471023,China)

机构地区：[1]河南科技大学,河南洛阳471023

出　　处：《计算机应用研究》2018年第7期2219-2221,2224,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61502364);河南省重点科技基金资助项目(132102210246);河南省产学研项目(142107000022);赛尔下一代网络创新项目(NGII20150311);河南科技大学博士科研启动基金资助项目

摘　　要：为了提高运动目标检测的准确度和精度,提出一种基于空时置信关系的运动检测方法。该方法利用快速核密度估计对图像像素点与其邻域像素点的空时关系进行建模,并根据样本值的离散度为背景模型分配对应的权重,最后依据像素值的背景隶属度权重均值,判断当前像素点属于运动前景还是背景。实验结果表明该方法对复杂场景具有更好的鲁棒性。This paper designed a novel scheme based on spatio-temporal confidence relationship to improve the performance of motion detection. This method employed fast kernel density estimation to model the relationship of a pixel and its neighbors,designated appropriate weight to the PDF（ probability distribution function） of environmental points according to the disperson degree of their sample values,and discriminated a target pixel belonging to background or foreground in a similar voting method by fusing the judging results of the environmental point set. Experiments on typical surveillance scenes verify that the proposed algorithm performs better than the state-of-the-art methods.

关 键 词：背景建模 目标检测 运动检测 运动分割 空时置信关系 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]