Max-Sum-Based Data Associations for Tracking Point and Extended Targets

For multitarget tracking applications, data association is a fundamental problem of assigning measurements to their corresponding targets. In this article, we propose two algorithms for tracking point and extended targets, respectively, based on factor graph representations of the joint probability...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE transactions on aerospace and electronic systems Ročník 61; číslo 2; s. 3059 - 3075
Hlavní autoři: Ma, Weizhen, Jing, Zhongliang, Dong, Peng, Leung, Henry
Médium: Journal Article
Jazyk:angličtina
Vydáno: New York IEEE 01.04.2025
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Aerospace and electronic systems
Algorithms
Approximation algorithms
Clustering
Clustering algorithms
Data association (DA)
Exact solutions
extended target tracking (ETT)
Extraterrestrial measurements
factor graph
Filters
Graph representations
Graphical representations
Heuristic algorithms
Kinematics
max-sum (MS) algorithm
Multiple target tracking
Normal distribution
point target tracking (PTT)
Probabilistic logic
Probability density function
Probability density functions
Simulation
Statistical analysis
Target tracking
Tracking
ISSN:0018-9251, 1557-9603
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:For multitarget tracking applications, data association is a fundamental problem of assigning measurements to their corresponding targets. In this article, we propose two algorithms for tracking point and extended targets, respectively, based on factor graph representations of the joint probability density functions. Both employ the max-sum (MS) algorithm to find the maximum a posteriori assignment such that the state of each target is updated with the most probable measurement(s). We model the single target densities as Gaussian distribution for point targets and gamma Gaussian inverse Wishart distribution for extended targets. Under linear Gaussian assumptions on the target models, the proposed algorithms provide analytical solutions to multitarget tracking problems. Specifically, the messages flowed in the factor graphs, existence probabilities and states of the targets are analytically calculated. These two algorithms have reduced computational load compared to the particle-based sum-product (SP) algorithms and avoid gating or clustering used by traditional multitarget tracking methods. We compare the proposed MS-based algorithms (MSAs) with the Poisson multi-Bernoulli mixture filters and the SP-based algorithms, and simulation results show that the MSAs have comparable or improved tracking performance.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9251
1557-9603
DOI:10.1109/TAES.2024.3482287