Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhanced Multi-Target Tracking in Dynamic Environments: Distributed Flooding Control in the Random Finite Set Framework

Aidan Blair, Amirali Khodadadian Gostar, Alireza Bab-Hadiashar, Xiaodong Li, Reza Hoseinnezhad

TL;DR

The paper addresses distributed multi-target tracking in dynamic environments for networks of mobile sensors by introducing a flooding-based distributed sensor control framework (DF-SC) that operates within the Random Finite Sets (RFS) paradigm. Each node runs an LMB filter, exchanges posteriors with neighbors, and coordinates actions via a flooding consensus mechanism to maximize information gain under practical constraints. A information-theoretic objective, approximated for tractability, guides the control decisions while enforcing collision avoidance and connectivity. Empirical results demonstrate that DF-SC achieves higher tracking accuracy and faster computation compared to baselines like fixed sensors and independent control, and offers competitive performance versus Distributed Coordinate Descent when accounting for computational cost. The approach has significant implications for scalable, real-time cooperative sensing in connected autonomous vehicle networks.

Abstract

Tracking multiple targets in dynamic environments using distributed sensor networks is a challenging problem for situational awareness in connected autonomous vehicles (CAVs). In such scenarios, the network of mobile sensors must coordinate their actions to accurately estimate the locations and trajectories of multiple targets, balancing limited computation and communication resources with multi-target tracking accuracy. Multi-sensor control methods can improve the performance of these networks by enabling efficient utilization of resources and enhancing the accuracy of the estimated target states. This paper proposes a novel multi-sensor control method that utilizes flooding-based communication to address this problem, ensuring distributed consensus of optimal sensor actions throughout the sensor network. Our method improves computational tractability and enables fully distributed control, ensuring the scalability and flexibility necessary for real-time CAV applications. Experimental results on several challenging multi-target tracking scenarios demonstrate that our approach significantly improves both multi-target tracking accuracy and computation time over competing methods.

Enhanced Multi-Target Tracking in Dynamic Environments: Distributed Flooding Control in the Random Finite Set Framework

TL;DR

The paper addresses distributed multi-target tracking in dynamic environments for networks of mobile sensors by introducing a flooding-based distributed sensor control framework (DF-SC) that operates within the Random Finite Sets (RFS) paradigm. Each node runs an LMB filter, exchanges posteriors with neighbors, and coordinates actions via a flooding consensus mechanism to maximize information gain under practical constraints. A information-theoretic objective, approximated for tractability, guides the control decisions while enforcing collision avoidance and connectivity. Empirical results demonstrate that DF-SC achieves higher tracking accuracy and faster computation compared to baselines like fixed sensors and independent control, and offers competitive performance versus Distributed Coordinate Descent when accounting for computational cost. The approach has significant implications for scalable, real-time cooperative sensing in connected autonomous vehicle networks.

Abstract

Tracking multiple targets in dynamic environments using distributed sensor networks is a challenging problem for situational awareness in connected autonomous vehicles (CAVs). In such scenarios, the network of mobile sensors must coordinate their actions to accurately estimate the locations and trajectories of multiple targets, balancing limited computation and communication resources with multi-target tracking accuracy. Multi-sensor control methods can improve the performance of these networks by enabling efficient utilization of resources and enhancing the accuracy of the estimated target states. This paper proposes a novel multi-sensor control method that utilizes flooding-based communication to address this problem, ensuring distributed consensus of optimal sensor actions throughout the sensor network. Our method improves computational tractability and enables fully distributed control, ensuring the scalability and flexibility necessary for real-time CAV applications. Experimental results on several challenging multi-target tracking scenarios demonstrate that our approach significantly improves both multi-target tracking accuracy and computation time over competing methods.
Paper Structure (22 sections, 1 theorem, 33 equations, 11 figures, 2 tables)

This paper contains 22 sections, 1 theorem, 33 equations, 11 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Statement and Background
  4. Overall architecture
  5. System components and model assumptions
  6. The multi-object filter
  7. Information fusion
  8. Distributed Information Fusion Challenges
  9. Distributed Multi-Sensor Control: The Proposed Solution
  10. Estimation
  11. Calculation of the predicted ideal measurement set (PIMS)
  12. Pseudo-update and fusion
  13. Objective Function
  14. Modified Kullback-Leibler Divergence
  15. Constraints
  16. ...and 7 more sections

Key Result

Lemma 4.1

$\text{If }\exists\,s\in\mathbb{S},1<t'<t\text{ such that }[\mathfrak{u}_s^*(t-1)=\mathfrak{u}_s^*(t'-1)]\land[\mathfrak{u}_s^*(t)=\mathfrak{u}_s^*(t')]\text{ then }\forall\,s'\in\mathbb{S},\mathfrak{u}_{s'}^*(t)=\mathfrak{u}_{s'}^*(t')$.

Figures (11)

  • Figure 1: Multiple UAVs must cooperatively inspect part of the ocean for monitoring of marine vehicles on the sea surface.
  • Figure 2: Overall operations running onboard each sensor node $s$, for tracking and control.
  • Figure 3: Diagram of a sensor with limited FoV, and the notation used for formulation of the probability of detection.
  • Figure 4: Probability detection variations against range for two cases in both of which $\rho_\min = 0$ and $\lambda=65\,\text{m}$, but maximum range is 600 m in one scenario and and 1000 m in another
  • Figure 5: The proposed architecture of the contents of multi-sensor control block executing at sensor node $s$, in the overall schematics shown in Figure \ref{['fig:overall_blockdiagram']}.
  • ...and 6 more figures

Theorems & Definitions (1)

  • Lemma 4.1