Sensor fusion cognition using belief filtering for tracking and identification

Erik Blasch, Lang HongProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 1999 · DOI

TL;DR

This paper introduces a "Cognitive Belief Filter" (CBF) that integrates sensor tracking and target identification for High Range Resolution (HRR) radar. By mimicking human cognitive fusion, it allows for dynamic estimation of the number of targets and their identities, even when the target count is unknown a priori, improving tracking accuracy in cluttered environments.

What Problem Does This Paper Address?

Accurately tracking and identifying multiple targets in complex, dynamic environments is a significant challenge, especially when the number of targets is not known beforehand. Traditional tracking algorithms often require a predefined number of targets, which can lead to errors if this assumption is incorrect. Furthermore, fusing kinematic (position, velocity) and feature-based (radar signatures) information for simultaneous tracking and identification is difficult, particularly when dealing with unknown or non-plausible targets. This paper aims to bridge this gap by developing a system that can dynamically infer the number of targets and their identities, leveraging a human-like cognitive fusion approach.

Methods

The core of the proposed solution is the Cognitive Belief Filter (CBF). This approach combines two main components:

  • Kinematic Tracking Belief Filter (TBF): This filter estimates the position and pose of targets using kinematic measurements. It employs a belief filtering approach to handle uncertainty and spurious measurements, allowing it to associate measurements with plausible targets.
  • Feature-based Classification Belief Filter (CBF): This component uses High Range Resolution (HRR) radar signatures to classify targets. It leverages statistical models of peak features (location and amplitude) within HRR profiles. The paper adapts the Statistical Feature based classifier (STaF) and uses a Bayesian approach to calculate class likelihoods and a posteriori probabilities. Crucially, it incorporates a confidence value calculation based on the cumulative distribution function (CDF) of in-class likelihoods to assess the reliability of a classification.

The cognitive fusion aspect comes into play by integrating the outputs of the TBF and CBF. The system presents the user with a set of plausible tracks and their associated classification confidences. The human operator then cognitively assesses this information to determine the correct number and identity of targets. This cognitive input then updates the tracking algorithm, refining the set of plausible targets and improving track quality. The paper also discusses belief evidence accrual using Dempster's rule of combination to fuse evidence from individual features and fused track and classified state estimation to combine kinematic and classification information.

Datasets

The paper utilizes the Moving and Stationary Target Acquisition and Recognition (MSTAR) program database. This database contains synthetic aperture radar (SAR) data at X-band with 1x1 foot resolution. The data includes images recorded at 15° and 17° depression angles with full aspect coverage. For this research, the SAR imagery was converted to HRR profiles. The authors trained the classifier on six known target classes (bmp2_apc, btr70_transport, ml tank, m2 apc, m109_gun, t72 tank) and left five classes as unknown. The simulation involved initiating tracks with an unknown number of targets and cluttered measurements.

Key Insights and Ideas

  • Mimicking Human Cognition for Fusion: The paper's central idea is that human cognitive abilities, particularly the ability to infer and adjust the number of targets, can be modeled and integrated into automated tracking systems. This "cognitive level fusion" is key to handling unknown target counts.
  • Confidence-Driven Decision Making: By explicitly calculating and using confidence measures for target classification, the system can avoid making erroneous decisions when faced with ambiguous or unknown targets. This is a significant improvement over purely Bayesian approaches that might force a decision.
  • Set-Theory Approach for Plausibility: The use of set theory to determine a "plausible set of targets" allows the system to manage uncertainty and provide a manageable set of hypotheses for the human operator to evaluate.
  • Synergy of Tracking and Identification: The CBF demonstrates that tightly integrating kinematic tracking with feature-based identification leads to more robust and accurate results than treating them as separate processes.

Key Results

The paper presents simulation results demonstrating the effectiveness of the CBF.

  • Improved Target Identification: The system successfully identified targets, as shown in the confusion matrix (Table 2), where it achieved reasonable belief scores for the actual target classes.
  • Dynamic Target Number Estimation: The simulations showed that the system could handle scenarios where the initial assumption of the number of targets was incorrect, allowing the user to correct it and improve tracking.
  • Faster Tracking: The CBF algorithm was observed to identify correct targets and assign measurements to tracks faster than conventional multitarget-multisensor tracking methodologies.
  • Handling Unknown Targets: The confidence-based approach allowed the system to manage unknown targets and avoid incorrect identifications, a critical capability for real-world scenarios.

Limitations and Future Work

  • Human-in-the-Loop Dependency: While the cognitive fusion is a strength, it implies a reliance on human operators for final decision-making, which might not be suitable for fully autonomous systems.
  • Computational Complexity: The paper mentions that the cognitive fusion approach may help control computational burdens, but the underlying belief filtering and statistical modeling can still be computationally intensive.
  • Unknown Target Statistics: The paper acknowledges the challenge of classifying unknown targets for which no prior statistics are available. While the confidence measure helps, robustly handling entirely novel targets remains an area for further research.

The authors suggest that the presented technique shows promise for multitarget tracking and warrants further exploration in problems with challenging environmental conditions such as effects, occlusions, lost sensor data, and unknown targets.

Summary updated

Abstract

Humans exhibit remarkable abilities to estimate, filter, predict, and fuse information in target tracking tasks, To improve track quality, we extend previous tracking approaches by investigating human cognitive-level fusion for constraining the set of plausible targets where the number of targets is not known a priori. The target track algorithm predicts a belief in the position and pose for a set of targets and an automatic target recognition algorithm uses the pose estimate to calculate an accumulated target-belief classification confidence measure. The human integrates the target track information and classification confidence measures to determine the number and identification of targets. This paper implements the cognitive belief filtering approach for sensor fusion and resolves target identity through a set-theory approach by determining a plausible set of targets being tracked.
BibTeX
@inproceedings{blasch1999sensor,
  author = {Erik Blasch and Lang Hong},
  title = {{Sensor fusion cognition using belief filtering for tracking and identification}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {1999},
  doi = {10.1117/12.341347},
}
Figure