扩展目标跟踪的集值运动建模与估计

Extended targets tracking has received increasing attention since the target kinematic state and the physical extension parameters are estimated simultaneously. This is beneficial for target identification and classification in practical applications. In order to describe the dynamics of the physical extension parameters in the tracking process, an extended target will be modeled as a combination of a time-varying number of sub-targets in this project, and a set-valued kinematic model will be proposed by using the random finite set approach, where each element consists of a random vector for target state and a random matrix for target extension. Then, the set-valued estimation methods will be proposed for tracking extended targets, in which the kinematic state constraints are incorporated between sub-targets. The multiple model set-valued filters will be proposed for tracking maneuvering extended targets when the jump Markov systems are used to describe target dynamics. For multi-sensor tracking problems, distributed set-valued estimation fusion approaches will be developed by determining exchanged information among neighboring sensors. The theoretic results will be verified by experiments involving tracking a mobile robot with arms. This project is of great significance in theoretical and practical applications since it not only develops novel methods for tracking extended targets but also provides solid theoretical guidance for set-valued estimations with state constraints.

扩展目标跟踪不仅估计目标的运动状态参数，而且估计目标的形状和大小等特征参数，有助于对目标的识别与分类，在目标跟踪领域受到广泛研究。为了更加准确地刻画跟踪过程中目标特征参数的变化，本项目将单个目标分解为时变数目的多个子目标，利用随机有限集方法建立扩展目标跟踪的集值运动模型，集合中的每个元素由描述子目标运动状态的随机向量和刻画子目标特征参数的随机矩阵构成。针对子目标之间固有的运动状态约束，以有限统计学理论为基础研究扩展目标跟踪的集值估计方法，并进一步研究建模为随机跳变系统的机动扩展目标跟踪，提出受运动状态约束的多模型集值估计方法。针对多传感器扩展目标跟踪，通过确定传感器之间交互的信息量及融合准则，提出分布式集值估计融合方法，并以带机械臂的移动机器人跟踪为应用对象进行仿真与实验验证。本项目的实施不仅能够提出新的扩展目标跟踪方法，而且对发展受状态约束的集值估计方法具有重要的科学意义与应用价值。

本项目围绕不同跟踪场景下的扩展目标运动建模与跟踪算法设计开展研究，取得的主要成果包括：建立了不同交互行为下多目标运动的非线性耦合模型，设计了交互行为切换、随机耦合、耦合强度不确定以及增益矩阵存在计算误差等不同跟踪场景下的非增广向量滤波算法，给出了确保跟踪误差有界的充分性条件；提出了基于组合测量和耦合测量下的多传感器分布式多目标协同跟踪算法，给出了确定测量增益矩阵的计算公式，分析了跟踪误差的有界性；提出了机动目标跟踪的交互式多模型矩阵估计方法和不同跟踪场景下的多传感器分布式滤波方法等，并进行了仿真与实验验证。项目组在国外期刊及国内外著名会议上发表论文30篇，其中SCI论文22篇，包括Automatica和IEEE汇刊8篇；申请发明专利5项，其中已授权2项；获2017年度教育部自然科学一等奖和第21届人工生命与机器人国际会议优秀论文奖。项目负责人于2016年受聘为国际SCI期刊IET Signal Processing编委。