基于概率数据关联与粒子滤波的多目标跟踪算法  

Study of Multiple Target Tracking Algorithm Based on Probabilistic Data Association and Particle Filter Technique

作　　者：王琳 寇英信[2] 于雷[2] 赵思宏[3] 

机构地区：[1]总参陆航研究所,北京101121 [2]空军工程大学工程学院,西安710038 [3]空军航空大学航理系,长春130022

出　　处：《系统仿真学报》2011年第11期2449-2454,共6页Journal of System Simulation

基　　金："十一五"国防预研基金项目(KJ-050402011);航空科学基金项目(20085196011)

摘　　要：提出一种基于概率数据关联和改进粒子滤波的多目标跟踪算法。该方法在分析关联区域内有效量测基础上,利用量测与目标预测位置的统计距离计算关联概率;然后,对粒子滤波器进行改进,将目标的最新量测和状态的高斯逼近组合在一起,对每个粒子采用类似于卡尔曼滤波的方式产生高斯建议分布,以此提高目标状态的估计精度;最后,将所有有效量测的估计结果按照关联概率进行加权,从而实现多目标跟踪。仿真结果表明,该算法用于复杂环境下的多目标跟踪精度较高,不仅降低了关联概率的计算难度,而且可以准确地进行数据关联,具有一定的工程应用价值。A multiple target tracking algorithm based on probabilistic data association and improved particle filter was proposed. On the basis of analysis effective observations in the association region, the association probability was calculated depending on the statistical distance between each measure and the predicted position of target. Then, the particle filter was improved. The new measurements of targets and Guass approach of targets state were combined together, and kalman filter was used to produce Guass advised distributing for each particle; At last, according to the association probability, all of the effective measures were added to make multiple target tracking come true. The results of simulation show the better performance and efficiency of multiple targets tracking algorithm in the clutter environment. Not only is the difficulty of association probability obviously reduced, but also the precision of data association is accurate. Besides these, the algorithm also shows the engineering application value.

关 键 词：概率数据关联 改进粒子滤波 多目标跟踪 关联概率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]