基于中智加权相似度量的尺度自适应视觉目标跟踪算法  

A scale adaptive visual object tracking algorithm based on weighted neutrosophic similarity coefficient

作　　者：胡珂立[1] 范恩[1] 叶军[1] 沈士根[1] 谷宇章[2] HU Keli;FAN En;YE Jun;SHEN Shigen;GU Yuzhang(Shaoxing University, Shaoxing 312000, China;Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China)

机构地区：[1]绍兴文理学院,浙江绍兴312000 [2]中国科学院上海微系统与信息技术研究所,上海200050

出　　处：《电信科学》2018年第5期50-62,共13页Telecommunications Science

基　　金：国家自然科学基金资助项目(No.61603258;No.61703280);浙江省公益技术应用研究项目(No.2016C31082)~~

摘　　要：中智集理论是传统模糊理论的拓展,它能够表述现实生活中的不确定性信息。面对不同问题,中智框架下的真(truth)、不确定(indeterminacy)、假(falsity)隶属度所占权重可能不同。提出了一种分量加权的余弦相似度量,并将其引入均值漂移视觉跟踪算法中。首先基于3σ理论和目标/背景相似度两种属性提出了相应的真、不确定、假量测,然后利用加权余弦相似度量构建权值向量,同时提出了基于中智加权余弦相似度量的尺度更新算法,综合提升均值漂移跟踪性能。实验结果表明,提出的视觉跟踪算法能较好克服相似背景、光照变化、尺度变化等挑战。The weight of the truth, indeterminacy, and falsity membership under the neutrosophic framework may be different when dealing with different problems. Due to this, a component weighted cosine similarity coefficient was proposed, and it was introduced into the mean shift tracking algorithm. Firstly, the corresponding methods for calculating the membership of the truth, indeterminacy, and falsity were proposed based on the theory of 3σ, as well as the similarity between the features of the corresponding area of the object and background. Then the weighted cosine similarity coefficient was used to construct the weight vector. In addition, a weighted cosine similarity coefficient based scale updating method was proposed. The experimental results demonstrate that the modified visual tracking algorithm performs well, even when there exists challenges like similar background, illumination or scale variation.

关 键 词：中智集 加权相似度量 目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]