摘要
本文以对算法实时性要求较高的实际场合为背景研究了集中式多传感器多目标跟踪问题,理论分析了该环境下现有经典算法的优缺点;在此基础上,研究了基于概率最近邻域算法的集中式多传感器多目标跟踪算法:提出了并行处理结构的集中式多传感器概率最近邻域算法,该算法首先利用概率最近邻域算法中最近邻量测源自目标的概率完成多目标数据互联,在各传感器送来的观测数据中获取一个与各目标航迹统计距离最小的量测,然后利用各个量测的统计距离以及其源于目标的概率将这些量测整合为一个等效量测,最后获得融合中心的状态估计;提出了顺序处理结构的集中式多传感器概率最近邻域算法,该算法利用离目标最近的量测源于目标的概率判断各传感器是否应参与融合中心的状态估计过程,降低了计算量。蒙特卡罗仿真结果表明:与经典的顺序多传感器联合概率数据互联算法相比,上述两种算法的耗时降低了50%以上,且在跟踪背景杂波适中的情况下能够有效跟踪目标。为了进一步验证上述两种算法的有效性和实用性,本文采用实测数据对算法进行验证,实测数据的处理结果表明在对算法实时性要求较高的实际场合下并行和顺序两种处理结构的集中式多传感器概率最近邻域算法的性能均优于顺序多传感器联合概率数据互联算法。
This paper studies the issue of centralized multisensor multitarget tracking in the occasions of high real-time requirement, and analyzes the advantages and disadvantages of existent classical algorithms theoretically under this environment. Based on the research and analysis, centralized multisensor multitarget tracking algorithms based on probabilistic nearest neighbor standard function algorithm are studied. A parallel centralized multisensor probabilistic nearest neighbor standard function algorithm is proposed. In this algorithm multitarget data association is achieved from the probabilities of the measurements with the shortest distance initiated from the targets, and the measurements with the shortest distance between the measurements from each sensor and each target track are obtained. This algorithm could form an equivalent measurement with these measurements and the probabilities of these measurements initiated from the targets, and get the state estimation of the fusion center finally. A sequential centralized multisensor probabilistic nearest neighbor standard function algorithm is also proposed. This algorithm makes use of the probability of the measurement with the shortest distance between the measurements and the target track to judge the participation of each sensor in the state estimation of the fusion center, which could reduce calculation burden. Monte Carlo simulation result shows that compared with classical sequential centralized multisensor joint probabilistic data association algorithm, the above mentioned two algorithms could track targets effectively under moderate clutter condition, and reduce the time consumption by more than 50%. The actual measured data was used to further testify the validity and practicability of the proposed algorithms, and test result shows that the performances of the above mentioned two algorithms are better than the performance of classical sequential centralized multisensor joint probabilistic data association algorithm in the occasions of high real-time requirement.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2010年第11期2500-2507,共8页
Chinese Journal of Scientific Instrument
基金
教育部新世纪优秀人才支持计划资助(批准号:NCET-05-0912)
国家自然科学(60672140)
全国优秀博士论文作者专项资金(200237)基金资助
关键词
集中式
概率最近邻域
多传感器多目标
性能分析
centralized structure
probabilistic nearest neighbor standard function
multisensor multitarget
performance analysis
