一种多特征自适应融合的粒子滤波红外目标跟踪方法  

A Particle Filter Infrared Target Tracking Method Based on Multi-feature Adaptive Fusion

作　　者：种衍文[1] 王泽文[2] 陈蓉[1] 王莹莹[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079 [2]杭州海康威视数字技术股份有限公司研究院,浙江杭州310051

出　　处：《武汉大学学报（信息科学版）》2016年第5期598-604,共7页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(41271398);中央高校基本科研业务费专项资金(2042014kf0242;2042014kf0263)~~

摘　　要：提出了一种自适应融合颜色特征和边缘特征的粒子滤波跟踪算法。首先,利用粒子滤波的天然框架,选择在红外条件下最能代表目标信息的颜色特征和边缘特征构造目标的多特征模型;然后,根据不同特征对目标与背景的可分性,对多特征模型中各特征分量的权值进行自适应调节;最后,借助动态空间模型,对粒子滤波跟踪算法进行改进,预测粒子的运动状态,从而克服环境突变对跟踪稳定性的影响。实验结果表明,本文算法能克服各种背景杂波及噪声的干扰,并能很好地解决目标在复杂背景下的尺度变化和突变运动带来的困难,保证了跟踪的鲁棒性和稳定性。In infrared target tracking it is difficult to predict the existence of background interference as there is a contradiction between the real-time and effectiveness of the algorithms. Hence this paper presents a particle filter tracking algorithm based on adaptive fusion of color features and edge features. Using the natural frame work of a particle filter, with infrared conditions it selects color features and edge features that best represent the target information to construct the multi-feature model of the target. According to the different feature separability of target and background, the weight of each feature component of a multi-feature is adaptively adjusted. This dynamic space model improves the particle filter tracking algorithm to predict the motion state of the particles to overcome the effects of environment mutations on tracking stability. Experimental results show that the proposed algorithm can overcome interference from all kinds of background clutter and noise ensuring tracking tobustness and accuracy

关 键 词：红外 多特征融合 粒子滤波 动态空间模型 目标跟踪 

分 类 号：P237.4[天文地球—摄影测量与遥感] TP391.4[天文地球—测绘科学与技术]