基于鉴别性与稳定性的自适应融合目标跟踪  

Discrimination and stability based adaptive fusion for target tracking

作　　者：严红平[1] 吕珂[1] 汪凌峰[2] 潘春洪[2] 

机构地区：[1]中国地质大学(北京)信息工程学院,北京100083 [2]中国科学院自动化研究所模式识别国家重点实验室,北京100190

出　　处：《计算机应用》2013年第A01期166-169,173,共5页journal of Computer Applications

基　　金：国家自然科学基金面上项目(61075016);中央高校基本科研业务费专项资金资助项目

摘　　要：提出了一种基于鉴别性与稳定性的自适应融合目标跟踪算法。在跟踪中,鉴别性度量目标与背景的区分程度,稳定性衡量跟踪框中心与目标实际中心之间的偏移程度。首先,对鉴别性与稳定性分开考虑,分别建模;而后将其引入自适应融合框架中,由此得到目标函数;最后优化目标函数得到自适应融合的权重。不同视频上的对比实验验证了该算法具有更高的跟踪准确性及稳定性。A target tracking method based on adaptive fusion was proposed, in which the fusion algorithm is derived from tracking discrimination and stability. In target tracking, discrimination measured the difference between target and background, while stability measured the deviation degree from true target center to tracking result. Algorithmically, discrimination and stability were modeled separately first. Then, they were introduced into the adaptive fusion framework, and further to formulate an object function. Finally, this object function was optimized to obtain the adaptive fusion weights. Comparative experiments on different kinds of videos show that the proposed algorithm holds higher tracking precision and stability.

关 键 词：自适应融合 目标跟踪 鉴别性 稳定性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]