Published in IET Radar, Sonar and Navigation
Received on 18th July 2011
Revised on 8th February 2012
doi: 10.1049/iet-rsn.2011.0262
ISSN 1751-8784
Evidence theory-based mixture particle filter for joint
detection and tracking of multiple targets
L. Jing
1
H. Yu
2
L. Feng
1
H. ChongZhao
1
1
MOE KLINNS Laboratory, Xi’an JiaoTong University, Xi’an 710049, People’s Republic of China
2
Schools of Aeronautics, Northwestern Polytechnical University, Xi’an 710072, People’s Republic of China
E-mail: elelj20080730@gmail.com
Abstract: In this study, a novel algorithm named evidence theory-based mixture particle filter is proposed for joint detection and
tracking for a varying number of targets in a cluttered environment. The posterior distribution of multiple target state considered in
single target state space is a multi-modal distribution with each mode corresponding to either a target or clutter. A general global
posterior distribution is adopted, which consists of existing components propagated from the previous time step, and new
components generated at the current time step to capture the newly appeared targets. An evidence theory-based framework is
utilised to determine the structure of the global posterior distribution. A set of masses are used to describe the possible kinds
of nature for each mixture component (e.g. it is from a target, clutter or undetermined at the current time step). The masses
are then transformed to a set of Pignistic probabilities, based on which a decision process is utilised to determine the nature
for each mixture component. The decision on the nature for each component (target or clutter) is made until sufficient
information arrives, which avoids the misjudgement because of insufficient information efficiently.
1 Introduction
Recently, joint detection and tracking for multiple targets has
been drawn much attention in the areas of surveillance radar
systems [1, 2], audio signal processing [3], wireless
communication [4] and so on. A lot of approaches have
been proposed to solve the joint detection and tracking
problem. Among them, multiple hypothesis tracking (MHT)
[5] is a multi-scan tracking algorithm, which maintains
multiple hypotheses associating past measurements with
targets. In the standard MHT algorithm, the number of
association hypotheses grows exponentially over time. In
recent years, the approaches based on joint multi-target
probability density (JMPD), which captures uncertainty
about the number of targets as well as their individual
states, are used widely in the joint detection and tracking.
Since JMPD is a high-dimensional entity that cannot be
computed in closed form, particle filters (PFs) have been
used to approximate the JMPD in realistic scenarios
involving tracking multiple targets [6]. Whereas the
PF-based JMPD approach is theoretically sound, it demands
intense computation, with a huge numbe r of particles
required to explore different dimensional state spaces for
target detection. More recently, a multi-scan Markov chain
Monte Carlo data association (MCMCDA) algorithm that
approximates the optimal Bayesian filter is proposed for
general multi-target tracking problems, in which unknown
numbers of targets appear and disappear at random times [7].
In the above methods, the dimension of the state vector is
proportional to the number of targets in the surveillance
region. They suffer from the curse of dimensionality
problem, since as the number of targets increases, the size
of the joint state space increases exponentially. In this
paper, the mixture PF [8] is adopted, in which the posterior
distribution of multiple target state is considered in single
target state space and approximated with a set of mixture
components. The mixture PF avoids the dimension problem
efficiently, and could handle the overlap and diffusion of a
number of targets via merging, splitting or reclustering the
particles generated at the initial stage. However, it could not
deal with the new target appearance problem since it could
not generate any new particles during the tracking process.
Moreover, the algorithm is utilised in a clutter-free
environment. In this paper, a general global posterior
distribution is adopted, which consists of the existing
components (including undetermined components and target
components) propagated from previous time step, and the
new components generated at the current time step to
capture the newly appeared targets. An evidence theory-
based framework is utilised to determine the structure of the
global posterior distribution via determining the nature for
each component (e.g. it is from a target, clutter or
undetermined at the current time step). The evidence theory
is an efficient tool to represent the uncertain knowledge
requiring fewer prior knowledge compared with the
Bayesian inference theory, and the detailed introduction to
evidence theory could be found in [9] .
Most of the multiple target tracking algorithms distinguish
target from clutter at each time step, which would probably
result in misjudgement when the information from sensors
is not sufficient to make a decision at the current time step.
In this work, an additional kind of nature, ‘undetermined’
IET Radar Sonar Navig., 2012, Vol. 6, Iss. 7, pp. 649–658 649
doi: 10.1049/iet-rsn.2011.0262
&
The Institution of Engineering and Technology 2012
www.ietdl.org
下载后可阅读完整内容，剩余9页未读，立即下载
weixin_38701952
- 粉丝: 5
- 资源: 977
我的内容管理
展开
最新资源
- 多功能HTML网站模板：手机电脑适配与前端源码
- echarts实战：构建多组与堆叠条形图可视化模板
- openEuler 22.03 LTS专用openssh rpm包安装指南
- H992响应式前端网页模板源码包
- Golang标准库深度解析与实践方案
- C语言版本gRPC框架支持多语言开发教程
- H397响应式前端网站模板源码下载
- 资产配置方案：优化资源与风险管理的关键计划
- PHP宾馆管理系统（毕设）完整项目源码下载
- 中小企业电子发票应用与管理解决方案
- 多设备自适应网页源码模板下载
- 移动端H5模板源码，自适应响应式网页设计
- 探索轻量级可定制软件框架及其Http服务器特性
- Python网站爬虫代码资源压缩包
- iOS App唯一标识符获取方案的策略与实施
- 百度地图SDK2.7开发的找厕所应用源代码分享
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
