Adaptive Interacting Multiple Model Algorithm Based on Information-Weighted Consensus for Maneuvering Target Tracking

Ziran Ding, Yu Liu, Jun Liu, Kaimin Yu, Yuanyang You, Peiliang Jing, You He

TL;DR

This paper introduces a novel algorithm, IMAM-UICF, for tracking maneuvering targets using a network of sensors. It combines an adaptive interacting multiple model approach with information-weighted consensus to improve tracking accuracy and sensor agreement, outperforming existing methods in complex scenarios.

What Problem Does This Paper Address?

Tracking maneuvering targets, especially in military applications like aircraft surveillance and radar tracking, is a significant challenge. Traditional tracking algorithms struggle when targets exhibit unpredictable maneuvers, leading to lost tracks and inaccurate estimations. This difficulty is amplified by the inherent uncertainty in target motion models. Furthermore, relying on a single sensor limits the overall system's robustness and accuracy. While multi-sensor systems offer improvements, effectively fusing data from distributed sensors while maintaining computational efficiency and robustness remains an open problem. This paper aims to address these issues by developing a distributed multi-sensor tracking system that can accurately estimate the state of maneuvering targets and ensure consistent estimates across all sensors.

Methods

The proposed algorithm, Adaptive Interacting Multiple Model Algorithm Based on Information-Weighted Consensus (IMAM-UICF), builds upon several key techniques:

  • Adaptive Interacting Multiple Model (IMAM): This approach uses a set of adaptive current statistical (ACS) models to better capture the target's varying motion characteristics, including its acceleration. This is an improvement over traditional Interacting Multiple Model (IMM) methods that use fixed models.
  • Unscented Transformation (UT): To handle the nonlinearities inherent in target motion and measurement models, the paper employs the Unscented Transformation. This technique accurately propagates the statistical characteristics of random variables through nonlinear functions, avoiding the linearization errors of Extended Kalman Filters (EKF) and offering better accuracy than Particle Filters (PF) for this application.
  • Information Filtering (IF): Instead of the standard Kalman filter, the paper utilizes an Information Filter. This is chosen because consensus protocols are shown to work more effectively with information filtering, particularly in distributed sensor networks.
  • Information-Weighted Consensus Protocol: This is the core of the distributed aspect. Each sensor node exchanges its estimated information (information matrix and vector) with its neighbors. Through iterative consensus updates, the estimates across the network converge, improving overall accuracy and agreement between sensors. The paper uses Metropolis weights for consensus, which are suitable for distributed architectures where nodes may not know the entire network topology.

Datasets

Not discussed in the paper. The paper uses simulated data for its experimental evaluation.

Key Insights and Ideas

  • Adaptive Models are Crucial for Maneuvers: The paper highlights that fixed motion models are insufficient for accurately tracking targets with complex maneuvers. Adaptive models, like those in IMAM, are essential for capturing these dynamic changes.
  • Information Filtering Enhances Consensus: The choice of Information Filtering as the base for each sensor's model is key. It allows for more effective information fusion through the consensus protocol, leading to better agreement among distributed sensors.
  • Consensus Bridges the Gap in Distributed Systems: In a distributed architecture, each sensor only has local information. The information-weighted consensus protocol allows sensors to indirectly fuse information from non-neighboring nodes, leading to estimates that are closer to the global optimal estimate, especially as the number of sensors increases.
  • UT Handles Nonlinearity Efficiently: The use of Unscented Transformation effectively addresses the nonlinear system dynamics without the computational burden or linearization errors associated with EKFs.

Key Results

  • Superior Tracking Accuracy: The proposed IMAM-UICF algorithm demonstrates superior tracking accuracy compared to IMAM, IMAM-DUIF, and IMM-UICF, particularly in scenarios with strong target maneuvers. This is evidenced by lower Position and Velocity ARMSE (Accumulative Root Mean Square Error).
  • Improved Sensor Agreement: The algorithm significantly reduces the disagreement between estimates from different sensors, achieving better consensus across the network. Figure 10 shows IMAM-UICF has lower disagreements than IMM-UICF.
  • Effectiveness with Varying Sensor Counts: As the number of sensors increases, the error of IMAM-UICF decreases, while IMAM-DUIF's error remains relatively unchanged, showcasing the benefit of the consensus protocol in leveraging more sensor data.
  • Robustness to Measurement Noise: IMAM-UICF maintains better estimation accuracy than the other methods even with increasing measurement noise.

Limitations and Future Work

  • Computational Cost: While IMAM-UICF offers improved accuracy and agreement, it incurs a higher computational cost than IMAM and IMM-UICF due to the iterative consensus operations and model adaptation. The paper notes that IMAM-UICF takes more time than IMAM-DUIF due to consensus and more time than IMM-UICF due to model adaptation.
  • Model Set Limitations: Although IMAM uses adaptive models, the paper implicitly acknowledges that the effectiveness of any multiple model approach depends on the chosen model set adequately representing the target's potential maneuvers.
  • Future Work: The paper doesn't explicitly detail future work, but the demonstrated effectiveness of IMAM-UICF suggests potential applications in more complex multi-target tracking scenarios, environments with sensor biases, or scenarios requiring real-time adaptation to unknown system dynamics. Further research could focus on optimizing the consensus weights or exploring more sophisticated adaptive model selection strategies.

Summary updated

BibTeX
@inproceedings{ding2018adaptive,
  author = {Ziran Ding and Yu Liu and Jun Liu and Kaimin Yu and Yuanyang You and Peiliang Jing and You He},
  title = {{Adaptive Interacting Multiple Model Algorithm Based on Information-Weighted Consensus for Maneuvering Target Tracking}},
  booktitle = {Sensors},
  year = {2018},
  url = {https://openreview.net/forum?id=dk2l6crOKP},
}
Figure