Flexible Fusion Structure-Based Performance Optimization Learning for Multisensor Target Tracking

Quanbo Ge, Zhongliang Wei, Tianfa Cheng, Shaodong Chen, Xiangfeng Wang2017 · DOI

TL;DR

This paper tackles the challenge of dynamically optimizing multisensor target tracking systems by introducing a "flexible fusion structure." It proposes a novel optimization model that balances tracking accuracy with system survivability, allowing the system to adapt its sensor configuration and fusion methods in real-time to changing operational conditions. This approach promises more robust and efficient target tracking in complex, dynamic environments.

What Problem Does This Paper Address?

Traditional multisensor target tracking systems often rely on fixed fusion structures, meaning the sensors involved and the fusion methods used remain constant throughout the tracking process. This rigidity is problematic in dynamic environments where factors like moving targets, non-cooperative adversaries, and limitations in communication bandwidth and measurement range can significantly degrade performance. The paper addresses the need for adaptive systems that can dynamically adjust their sensor configuration and fusion strategies to maintain optimal performance under these varying conditions. Specifically, it aims to solve the problem of "sensor subsets selection" – deciding which sensors should participate in centralized and distributed fusion processes at any given time.

Methods

The core of this paper is the design of a flexible fusion algorithm that leverages optimization learning technology. The authors introduce two key performance indexes:

  • Tracking Accuracy: Measured by the trace of the fusion estimation error covariance. A lower trace indicates better accuracy.
  • System Survivability: Defined as the system's ability to avoid detection by non-cooperative targets. This is formulated based on data communication traffic, where less communication implies higher survivability.

Based on these indexes and considering constraints like communication bandwidth and measurement distance, the paper establishes optimization models for both single-target and multi-target tracking. A key innovation is the formulation of a flexible fusion structure that allows for dynamic adjustment of sensor subsets.

For solving these optimization models, the paper proposes a step-by-step solution strategy to manage the combinatorial complexity, especially for multi-target scenarios. This strategy involves progressively narrowing down the sensor subsets based on constraints. For multi-target allocation, the Ant Colony Optimization Algorithm (ACOA) is employed to determine the optimal assignment of sensor nodes to targets.

Datasets

Not discussed in the paper. The simulations are based on synthetic data generated to represent specific scenarios (e.g., enemy and friendly plane positions, target states, noise characteristics).

Key Insights and Ideas

  • Survivability as a Critical Performance Metric: While tracking accuracy is standard, the paper highlights the importance of system survivability, especially in military or adversarial contexts. This dual-objective optimization is a significant contribution.
  • Dynamic Sensor Subset Selection: The concept of a "flexible fusion structure" that can dynamically reconfigure sensor participation in fusion processes is crucial for adapting to real-world complexities.
  • Trade-off Between Accuracy and Survivability: The paper implicitly and explicitly discusses the inherent trade-off: increased sensor participation (for accuracy) often leads to higher communication traffic (reducing survivability). The optimization model aims to find the best balance.
  • Step-by-Step Optimization for Scalability: The proposed solution strategy for complex multi-target scenarios is essential for making the optimization tractable, avoiding the "combinatorial explosion" problem.
  • ACOA for Multi-Target Allocation: The application of ACOA demonstrates its effectiveness in solving discrete combinatorial problems like sensor node allocation in a multi-target environment.

Key Results

The paper demonstrates the feasibility and effectiveness of its proposed flexible fusion structure and optimization model through simulations:

  • Single Target Tracking: In a scenario with a potential node failure, the system successfully adapted by reallocating resources and maintaining performance. The optimized system achieved an estimated trace of the state estimate error covariance of approximately 2.6026 and a survivability index of 0.9541.
  • Multi-Target Tracking: In a scenario involving two targets and a potential security risk constraint violation for one node, the system again adapted. The optimal allocation resulted in estimated error covariances of approximately 5.6625 for target T2 and 7.1698 for target T3, with a survivability index of 0.8909.
  • Improved Self-Adaptation: The simulations show that the system can effectively handle issues like node failures, measurement distance constraints, and security risk constraints, demonstrating its self-adaptive capabilities.

Limitations and Future Work

  • Limited Scope of Survivability: The paper primarily focuses on data communication traffic as the sole factor influencing survivability. Other factors like radar radiation are acknowledged but not modeled.
  • Specific Fusion Methods: While the optimization framework is general, the simulations primarily use the Unscented Kalman Filter (UKF) and specific fusion methods (Dimension Expansion and Local Estimate Weighted Fusion). Future work could explore integrating other advanced nonlinear filters.
  • Real-World Data: The simulations rely on synthetic data. Validation with real-world datasets would be a crucial next step.
  • Generalization of Survivability Index: The paper proposes a specific third-order inertia logarithmic function for survivability. Exploring and comparing other formulations for the survivability index is suggested for future work.
  • More Complex Scenarios: Extending the model to handle more complex scenarios, such as dynamic threats, jamming, or more intricate network topologies, would be valuable.

Summary updated

BibTeX
@misc{ge2017flexible,
  author = {Quanbo Ge and Zhongliang Wei and Tianfa Cheng and Shaodong Chen and Xiangfeng Wang},
  title = {{Flexible Fusion Structure-Based Performance Optimization Learning for Multisensor Target Tracking}},
  year = {2017},
  doi = {10.3390/s17051045},
}
Figure