文档介绍：摘 要
本文主要对多目标跟踪中的数据关联算法进行了研究。数据关联是机动目标
跟踪中的一项关键技术。本文在概述了目标跟踪的基本理论及基本环节基础上，
对数据关联特点和过程进行了全面的分析和概括总结，并对数据关联算法中的几
种经典算法：最近邻算法、概率数据关联算法、联合概率数据关联算法和交互多
模型概率数据关联算法进行了深入的研究。在高噪声密集杂波环境下，仅利用目
标的运动量测信息，很难将目标和杂波区分开。针对这一问题，本文融入目标的
特征量测信息，创新性的提出了一种基于最小二乘的多特征概率数据关联 EM 方
法，利用多个特征信息，增强目标和杂波的区分能力。仿真结果表明，该方法能
够增强区分目标和杂波的能力，并能够减小相近特征量测所引起的跟踪误导，显
著地提高对目标的跟踪精度。此外，在航迹关联中，本文对 Viterbi 数据关联算法
进行了改进。在 Viterbi 数据关联算法基础上，结合最近邻思想，进行二次关联，
减小虚假航迹的干扰，对目标航迹进行最优选择。这种改进的 Viterbi 数据关联方
法能够连续处理丢失的量测及跟踪目标，提供一系列最好的跟踪目标集，并能够
减小检测概率和杂波密度对跟踪性能的影响。仿真实验验证了该方法在低检测概
率密集杂波环境下的优越性。最后，对论文研究工作进行了回顾总结，并对今后
的研究工作进行了展望。
关键词：数据关联，期望极大化（EM），特征量测，二次关联，Viterbi 算法
I
ABSTRACT
The data accociation algorithms of multi-target tracking are studied.
Firstly, based on the overview of fundamental theories and essential links of
multi-target tracking, data association is introduced as a key technique of maneuvering
target tracking, besides, its characteristics and processing procedures are overall
analysized and generalized. Moreover, some of basic data association algorithms,
including nearest neighbor data assciation(NNDA), probability data assciation(PDA),
joint probability data assciation(JPDA), and interactive multi-model algorithm(IMM),
are studied deeply.
Secondly, in the environment of high noise and high-density clutter, it’s difficult to
discriminate targets from clutter, just using the kinematic information. Aiming at this
problem, an expectation maximization algorithm for multi-feature aided probabilistic
association based on least square, that is incorporated targets’ feature information into
the tracking process, is proposed in the thesis. The new algorithm can improve the
capability of tracking system between targets and clutter. It’s shown by simulating
results that the