一种基于证据推理的视频运动目标融合检测算法  

A fusion approach for video moving object detection based on evidential reasoning

作　　者：韩德强[1,2] 程博[1,2] 杨艺[3,4] 

机构地区：[1]西安交通大学智能网络与网络安全教育部重点实验室,西安710049 [2]西安交通大学电子与信息工程学院,西安710049 [3]西安交通大学机械振动与强度国家重点实验室,西安710049 [4]西安交通大学航天航空学院,西安710049

出　　处：《控制与决策》2016年第8期1345-1353,共9页Control and Decision

基　　金：国家973计划项目(2013CB329405);国家自然科学基金项目(61573257;61203222);陕西省科技计划项目(2013KJXX-46);教育部博士点基金项目(20120201120036);中央高校基本科研业务费专项资金项目(xjj2012104;xjj2014122)

摘　　要：为了解决复杂场景下传统的运动目标检测问题,利用证据推理—–谨慎有序加权平均方法(COWA-ER),提出一种综合使用混合高斯、均值滤波和码本的多方法融合的检测方法.该融合检测算法以上述3种检测方法为准则建立一个多准则决策框架,通过双阈值检测法来表征检测过程中的不确定性,最终利用COWA-ER方法进行决策级融合,实现多种方法的优势互补.实验表明,所提出的融合检测算法具有更理想的目标检测效果,能有效应对诸如阴影及光照突变等问题对检测性能的影响.To handle the problems encountered in traditional moving object detecting algorithms under complex scenarios,by comprehensively using the Gaussian mixture model（GMM）, mean filter and codebook（CB）, a fusion-based detection approach is proposed based on cautious ordered weighted averaging with evidential reasoning（COWA-ER）. In the proposed approach, a multi-criteria decision-making framework is established, where the three detection algorithms are used as multiple criteria. The double threshold method is used to model the uncertainty in the detection. The decision-level fusion for detection is finally accomplished by using COWA-ER, where complementary advantages of the three algorithms can be fully used. Experimental results show that the proposed approach can achieve better performance for the detection, and effectively deal with the influence of detection performance caused by shadow and light change.

关 键 词：视频目标检测 不确定性 COWA-ER 融合检测 多准则决策 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]