融合多特征的粒子滤波目标跟踪算法  

Particle filter targets tracking by fusing multiple features

作　　者：牛长锋[1] 刘玉树[1] 

机构地区：[1]北京理工大学智能信息技术北京市重点实验室,北京100081

出　　处：《华中科技大学学报（自然科学版）》2010年第1期18-21,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国防"十一五"预研项目

摘　　要：针对在复杂环境下,利用单个特征不能准确跟踪目标的问题,提出了一种在粒子滤波框架下融合多个特征进行目标跟踪的算法.利用Fisher线性判别原理,从特征集合中抽取能有效判别目标和背景并能保持目标时域一致的特征子集.在粒子滤波框架下,自适应地融合特征子集中的所有特征实现目标跟踪.通过试验证明了该算法在遮挡、环境变化等情况下的健壮性,同时也验证了跟踪结果的精确性.Under complex environment, it is impossible to track target accurately by only a feature. Thus, a novel object tracking method was proposed by fusing multi-feature in the framework of particle filter. Using Fisher linear discriminative theory, some discriminative features were extracted from the set of features, by which target could be distinguished from background very well and target appearance was kept consistent on time domain. Then, all discriminative features were fused adaptively in the framework of particle filter to track target. Experiment results show the proposed method is robust to occlusion and environmental change with high precision.

关 键 词：目标跟踪 粒子滤波 特征融合 特征选择 Fisher线性判别原理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]