A Sequential Two-Stage Track-to-Track Association Method in Asynchronous Bearings-Only Sensor Networks for Aerial Targets Surveillance

Yu Yang, Qingyu Hou, Wěi Zhāng, Jinxiu ZhangSensors 2019 · DOI

TL;DR

This paper introduces a novel sequential two-stage track-to-track association (TTTA) method for asynchronous bearings-only sensor networks. It addresses the challenge of associating targets when sensor data isn't perfectly synchronized, a common issue in real-world surveillance, by proposing a new likelihood function that leverages joint temporal and spatial constraints.

What Problem Does This Paper Address?

In aerial target surveillance using sensor networks, a crucial task is Track-to-Track Association (TTTA). This involves correctly grouping individual tracks (mono tracks) from different sensors that originate from the same target. A major hurdle in practical scenarios is that sensors often operate asynchronously, meaning their data isn't collected at the exact same time. This asynchronicity, coupled with potential sensor biases and missed detections, complicates the process of determining if two tracks belong to the same target. Existing TTTA methods often assume synchronized data or rely on simplified geometric constraints (like hinge angle differences) that can fail in certain target configurations. This paper tackles the problem of performing accurate TTTA in asynchronous bearings-only sensor networks, where only angle measurements are available, without requiring data synchronization.

Methods

The core of the proposed method, termed Joint Temporal and Spatial Constraints (JTSC), is a novel likelihood function designed to evaluate the similarity between asynchronous mono tracks. This likelihood is built upon a sequential two-stage strategy:

  1. Pairwise Fusion Model: This stage takes two asynchronous mono tracks from different sensors and fuses them to estimate a potential 3D trajectory (a "stereo track"). This is achieved using an "asynchronous crossing-location" approach, which estimates the target's position by finding the midpoint of the common perpendicular between the lines of sight (LOS) from the sensors to the target, considering a time interval threshold.
  2. Pairwise Similarity Model: This stage evaluates the similarity between two stereo tracks generated in the first stage. It identifies "clues" or Joint Temporal-Spatial Matching Points (JTSMPs) – pairs of data points from different stereo tracks that are close in both time and space. A decaying function quantifies the strength of these clues, and a "score of clues" is calculated for each point. The overall pairwise similarity is then an average of these scores, providing a measure of how likely two stereo tracks represent the same target.

The overall tuple likelihood (representing the probability of multiple tracks belonging to the same target) is then computed as a product of these sequential pairwise similarities. This approach avoids the need for global synchronization and directly handles asynchronous measurements.

Datasets

The paper uses simulated data for evaluation. The scenario involves three maneuvering aerial targets and three stable bearings-only sensors deployed in a 3D space. The simulations are run 100 times using Monte Carlo methods to assess performance under various conditions. No specific real-world datasets are used for training or evaluation.

Key Insights and Ideas

  • Asynchronous Data is a Feature, Not Just a Bug: Instead of trying to synchronize asynchronous data (which can be computationally expensive and introduce errors), the JTSC method directly leverages the asynchronous nature of measurements.
  • 3D Geometry Matters: By fusing bearing-only measurements into estimated 3D positions (stereo tracks), the method captures richer geometric information than methods relying solely on 2D angles. This is crucial for distinguishing targets in complex scenarios.
  • Sequential Processing for Scalability: The two-stage sequential approach allows for processing multiple sensors without requiring a global synchronization step, making it more scalable for larger sensor networks.
  • Temporal and Spatial Constraints are Key: The joint use of temporal and spatial constraints in identifying matching points (JTSMPs) and evaluating similarity is central to the method's robustness against noise and asynchronous sampling.

Key Results

  • The proposed JTSC method demonstrates favorable association performance without data synchronization.
  • Simulations show that JTSC outperforms methods based on Hinge Angle Differences (HADs), particularly in scenarios where targets might have similar HADs but distinct 3D trajectories.
  • The performance of JTSC is analyzed across various factors:
  • Association Time: JTSC's performance improves significantly with longer track histories, surpassing HADC and NN after about 20 seconds.
  • Target Density (Offset Time): JTSC shows robust performance and grows faster than HADC and NN as target density increases.
  • LOS Errors: JTSC's performance degrades more sharply with increasing Line-of-Sight (LOS) errors compared to HADC and NN, as its pairwise fusion relies on accurate position estimation.
  • Sensor Deployment: JTSC is sensitive to sensor geometry, performing best with a wider apex angle (around 90°) and less well with narrow angles.
  • The method's effectiveness is validated by comparing the probability of correct association (Pca) across different parameters and conditions.

Limitations and Future Work

  • Sensitivity to LOS Errors: The paper acknowledges that JTSC's performance deteriorates significantly with larger LOS errors, as the pairwise fusion step is sensitive to these inaccuracies.
  • False Tracks: The authors note that while their method reduces false tracks, they are still a challenge that needs to be addressed.
  • Future Work: The authors suggest exploring modifications to normalize the likelihood for consistency and investigating advanced Artificial Intelligence approaches to further improve effectiveness and efficiency. They also mention the need to balance effectiveness and computational efficiency.

Summary updated

Abstract

Successful track-to-track association (TTTA) in a multisensor and multitarget scenario is predicated on a reasonable likelihood function to evaluate the similarity of asynchronous mono tracks. To deal with the lack of synchronous data and prior knowledge of the targets in practical applications, this paper investigates a global optimization method with a novel likelihood function constructed by finite asynchronous measurements with joint temporal and spatial constraints (JTSC). For a scenario with more than two independent sensors, a sequential two-stage strategy is proposed to calculate the similarity of multiple asynchronous mono tracks. For the first stage, based on the temporal features of measurements from different sensors, a pairwise fusion model to estimate the position of the target with two mono tracks is established based on the asynchronous crossing location approach. For the other stage, to evaluate the similarity of the outputs, a pairwise similarity model is constructed by searching for the optimal matching points by setting temporal and spatial constraints. Thus, the likelihood of multiple asynchronous tracks is obtained. Simulations are performed to verify that the proposed method can achieve favorable performance without data-synchronization, and also demonstrate the superiority over the methods based on hinge angle differences (HADs) in some scenarios.
BibTeX
@inproceedings{yang2019sequential,
  author = {Yu Yang and Qingyu Hou and Wěi Zhāng and Jinxiu Zhang},
  title = {{A Sequential Two-Stage Track-to-Track Association Method in Asynchronous Bearings-Only Sensor Networks for Aerial Targets Surveillance}},
  booktitle = {Sensors},
  year = {2019},
  doi = {10.3390/s19143185},
}

Loading related content...

Figure