Multiple Vessel Cooperative Localization Under Random Finite Set Framework With Unknown Birth Intensities

Feihu Zhang, Le Li, Lichuan Zhang, Guang PanIEEE Access 2020 · DOI

Authors: Feihu Zhang, Le Li, Lichuan Zhang, and Guang Pan
Affiliation: School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an, China
Journal: IEEE Access

Introduction

Cooperative localization (CL) is a critical capability for maritime operations, enabling a fleet of vessels to achieve higher positional accuracy than individual ships could alone. By sharing proprioceptive data (e.g., GPS, IMU) and exteroceptive measurements (e.g., radar, sonar), a group of agents can mitigate the errors inherent in standalone navigation. However, real-world maritime environments present significant challenges, particularly regarding data association in cluttered waters and the unpredictable appearance of new vessels within a surveillance area.

System architecture for cooperative localization among three vessels utilizing radar, GPS, and inter-vessel data exchange.

The Data Association Challenge

In dense maritime traffic or high-clutter environments, traditional localization methods struggle with the "chicken-and-egg" problem of data association: accurate position estimates are required to correctly associate radar measurements with specific vessels, yet correct association is a prerequisite for accurate estimation. Furthermore, standard Bayesian filters, such as the Probability Hypothesis Density (PHD) filter, typically require a priori knowledge of "birth intensities"—pre-defined spatial regions where new targets are likely to appear. In dynamic maritime scenarios, such knowledge is rarely available, making it difficult to detect and track vessels that enter the fleet's sensor range unexpectedly.

Proposed Methodology

This paper introduces an improved CL algorithm that leverages the Random Finite Set (RFS) framework to bypass explicit measurement-to-target association. The core of the approach is a Gaussian Mixture PHD (GM-PHD) filter enhanced by a novel point-matching pattern recognition technique.

1. RFS and the GM-PHD Filter

By treating the fleet and its measurements as Random Finite Sets, the system propagates the "intensity" of targets across the state space rather than tracking individual identities. This naturally accommodates a varying number of targets and suppresses clutter without the computational exponentiality of traditional association algorithms.

2. Topology-Based Birth Intensity Estimation

The primary innovation of this work is the elimination of pre-defined birth intensities. The authors propose a point-matching algorithm based on the observation that the spatial topology (relative distances and geometry) of a vessel group remains relatively invariant over short time intervals. When new measurements appear that do not align with existing tracks but do conform to the group's known topology, the system classifies them as "new births" rather than random clutter.

3. Two-Step Update Process

The filter operates in two distinct stages:

  • Proprioceptive Update: The filter is first updated using the ship's internal navigation data.
  • Exteroceptive Update: A second update is performed using radar measurements from other vessels. Measurements that match the invariant topology are used to initialize new Gaussian components in the PHD filter, allowing for the seamless integration of new targets.

Experimental Validation

The proposed framework was validated using simulated maritime environments ($1000 \times 1000$ m) under both Constant Velocity (CV) and Constant Turn (CT) motion models. Sensors were modeled with realistic noise profiles (GPS: $25m^2$; Radar: $4m^2$).

Key Results

  • Detection of Unknown Births: In scenarios where three new vessels appeared at staggered intervals, the proposed method successfully identified and tracked them. In contrast, standard PHD filters without birth intensity information failed to detect these vessels, dismissing them as clutter.
  • OSPA Performance: Using the Optimal Subpattern Assignment (OSPA) metric, the algorithm demonstrated a stable and low error rate. While standard filters showed significant error spikes upon the arrival of new vessels, the proposed method maintained tracking continuity.
  • Bandwidth Efficiency: The RFS-based approach proved more efficient than traditional state-exchange methods (like SECL), saving approximately $20n^2$ bits of communication bandwidth by avoiding the transmission of large covariance matrices for every potential target-measurement pair.

Conclusion

By combining the robustness of the RFS framework with the geometric intelligence of point-matching, this research provides a scalable solution for multi-vessel cooperative localization. The ability to estimate birth intensities dynamically allows the system to operate in unpredictable environments without manual tuning. Future work will focus on extending this framework to handle temporary sensor dropouts through curve-fitting and smoothing techniques, as well as validating the algorithm against real-world maritime datasets.

Summary updated

BibTeX
@article{zhang2020multiple,
  author = {Feihu Zhang and Le Li and Lichuan Zhang and Guang Pan},
  title = {{Multiple Vessel Cooperative Localization Under Random Finite Set Framework With Unknown Birth Intensities}},
  journal = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2019.2963445},
}
Figure