一种新的证据融合方法及在目标跟踪中的应用  

A New Fusion Method of Evidence and its Application in Target Tracking

作　　者：曹洁[1] 李晓瑶[1] 

机构地区：[1]兰州理工大学计算机与通信学院,兰州730050

出　　处：《小型微型计算机系统》2015年第4期891-896,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61263031)资助

摘　　要：为了有效融合高度冲突的证据,本文在余弦相似度和K-L距离基础上提出一种证据自适应融合方法.首先联合余弦距离和经典冲突系数定义了一种新的证据冲突衡量标准;然后利用K-L距离获得待组合证据的权重,提出一种基于序贯式的证据组合方法.而后考虑到Dempster组合规则可以有效融合低冲突证据,为使新的组合方法和Dempster组合规则发挥各自的优势,提出了一种组合规则自适应切换的融合方法.最后基于AMI语料库对本文融合方法的有效性进行了验证.实验结果表明:相比传统的证据理论融合方法,本文方法具有较好的准确性和稳定性,可满足视频跟踪的应用需求.To fuse the high conflict evidences, a method that adaptive evidence combination based on cosine similarity and K-L dis- tance is proposed in this paper. Firstly, define a novel evidence conflict representation jointed classical conflict factor through cosine distance. Besides, based on this definition, use K-L distance to obtain the weight of evidence, and propose the sequential combination of rules. Then,considering the Dempster＇s combination rules can effectively fuse the low conflict evidence with each other. In order to make most use of the advantages of Dempster＇s combination rules and proposed algorithrn,a novel fusion strategy is proposed which can adaptively switch the combination rule. Lastly,the experiment which is based on AMI corpus is made to test and verify the pro- posed algorithm efficiency. It is showed that the algorithm has better accuracy and stability, and can meet the requirement of video tracking.

关 键 词：计算机视觉 DEMPSTER组合规则 多特征自适应融合 余弦距离 K—L变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]