arxiv: 2604.13884 · v1 · submitted 2026-04-15 · 📡 eess.SP

Recognition: unknown

A Variational Message Passing Framework for Multi-Sensor Multi-Object Tracking using Raw Radar Signals

Anders Malthe Westerkam , Jakob M\"oderl , Erik Leitinger , Troels Pedersen

Authors on Pith no claims yet

Pith reviewed 2026-05-10 12:49 UTC · model grok-4.3

classification 📡 eess.SP

keywords multi-object trackingvariational message passingraw radar signalsmulti-sensor fusionBernoulli-Gamma modelUAV surveillancedirect trackingmean-field approximation

0 comments

The pith

A variational message passing algorithm performs multi-object tracking directly on raw signals from multiple radars, handling an unknown number of weak and closely spaced targets more effectively than conventional pipelines.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a variational message passing framework for tracking an unknown number of objects directly from raw radar signals received by multiple sensors. It uses a superimposed signal model and a hierarchical Bernoulli-Gamma distribution to jointly handle object existence, reflectivities, and sensor reliability. This approach avoids the information loss in conventional detect-then-track methods that rely on pre-processed measurements. In simulations with closely spaced weak objects, it shows improved performance over super-resolution estimation followed by belief propagation tracking, especially in low signal-to-noise and high clutter conditions. The method performs detection, tracking, state estimation, and parameter learning in one unified inference process.

Core claim

The authors develop a direct multi-object tracking method based on variational message passing that processes raw MIMO multi-radar signals jointly. By modeling the signals as a superposition from multiple objects and using a hierarchical model for existence and amplitudes, the VMP algorithm derives efficient message updates under mean-field approximation. This allows simultaneous object detection, track formation, and state estimation while learning nuisance parameters, outperforming pipelines that separate super-resolution estimation from tracking in challenging scenarios.

What carries the argument

The variational message passing (VMP) algorithm operating on the mean-field approximated joint posterior over object states, existence indicators, reflectivities, and link reliabilities in the hierarchical Bernoulli-Gamma model.

If this is right

Allows coherent data fusion across multiple radar sensors by joint signal processing.
Handles unknown and time-varying number of objects without explicit association steps.
Maintains performance for closely-spaced objects by accounting for signal correlations in the raw data.
Provides online learning of reflectivities and other parameters during tracking.
Improves robustness in low-SNR and clutter-rich environments for applications like UAV surveillance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This framework could extend to other sensor types where raw signal models are available, such as sonar or lidar.
If the mean-field approximation holds in real data, it suggests variational methods can replace multi-stage pipelines in sensor fusion tasks.
The approach implies potential reductions in false tracks by directly modeling existence probabilities from signals.
Testing on measured radar data with actual drones would validate the simulation advantages in practical settings.

Load-bearing premise

The mean-field approximation used to derive the message updates accurately represents the dependencies in the multi-object tracking posterior.

What would settle it

A simulation or experiment where the direct VMP method is applied to raw signals from closely spaced objects at very low SNR and its tracking accuracy is compared to the conventional super-resolution plus BP method; if the conventional method performs better or equally, the claim would be falsified.

Figures

Figures reproduced from arXiv: 2604.13884 by Anders Malthe Westerkam, Erik Leitinger, Jakob M\"oderl, Troels Pedersen.

**Figure 2.** Figure 2: Bayesian network representation of the multi-object detection and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Tracks of the considered scenarios. (a) The random crossing tracks [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 6.** Figure 6: The OSPA eror for the sensor handover track. The results are averaged [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗

**Figure 5.** Figure 5: The OSPA error for the parallel tracks maneuver. The results are [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

read the original abstract

The growing proliferation of unmanned aerial vehicles (UAVs) poses major challenges for reliable airspace surveillance, as drones are typically small, have low radar cross-sections, and often move slowly in cluttered environments. These characteristics make the joint tasks of detecting, localizing, and tracking multiple objects difficult for conventional detect-then-track (DTT) approaches, which rely on pre-processed measurements and may discard informative low-signal-to-noise ratio (SNR) signal components. To overcome these limitations, we propose a variational message passing (VMP)-based direct multiobject tracking (MOT) method that operates directly on raw radar signals and explicitly accounts for an unknown and time-varying number of objects. The proposed method is formulated for MIMO multi-radar systems and performs data fusion by jointly processing the signals of all radar sensors using a probabilistic model. A superimposed signal model is employed to capture correlations in the raw sensor data caused by closely spaced objects, and a hierarchical Bernoulli-Gamma model is introduced to jointly model object existence, reflectivities, and the reliability of individual radar-object links. Using a mean-field approximation, we derive message updates, yielding a computationally efficient VMP algorithm that simultaneously performs object detection, track formation, state estimation, and nuisance parameter learning directly from the radar signal. Simulation results in synthetic scenarios with weak and closely-spaced objects show that the proposed direct-MOT method outperforms a conventional pipeline based on super-resolution estimation followed by belief propagation (BP)-based tracking, particularly in low-SNR and clutter-rich conditions, demonstrating the advantages of direct signal-level inference and coherent multi-radar fusion.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a variational message passing framework for direct multi-object tracking from raw multi-radar signals, with simulations showing gains but mean-field assumptions potentially limiting for close targets.

read the letter

The paper gives a variational message passing framework for doing multi-object tracking straight from raw radar signals in a multi-radar setup. The simulations look better than the usual pipeline, but the mean-field assumption needs scrutiny for close objects. They introduce a hierarchical Bernoulli-Gamma model to tie together object existence, reflectivities, and link reliability. The superimposed signal model captures correlations from nearby targets, and they derive efficient message updates for joint detection, tracking, and parameter learning. This direct approach avoids losing information in pre-processing, which is useful for low-SNR UAV detection in clutter. The results in synthetic scenarios with weak and close objects show advantages in low SNR and clutter. That part is clear from the abstract. The potential issue is that mean-field factorization decouples the object states, which might not hold when signals overlap heavily. If the paper doesn't check this with tighter bounds or alternative approximations, the performance claims could be optimistic. Also, no mention of real measurements or statistical significance tests. This paper is for specialists in radar signal processing and multi-target tracking who are interested in variational methods. A reader working on similar fusion problems would get value from the model construction and the comparison. It deserves a serious referee because the problem is timely and the method is a coherent extension of existing tools. The authors should be asked to address the approximation validity and add more evidence.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a variational message passing (VMP) algorithm for direct multi-object tracking (MOT) from raw MIMO multi-radar signals. It employs a superimposed signal model to capture correlations from closely spaced objects and introduces a hierarchical Bernoulli-Gamma model to jointly represent object existence, reflectivities, and per-link reliability. Under a mean-field approximation, message updates are derived that simultaneously perform detection, track formation, state estimation, and nuisance-parameter learning. Simulations in synthetic low-SNR, clutter-rich scenarios with weak, closely spaced targets report that the direct-MOT method outperforms a conventional super-resolution estimation followed by belief-propagation tracking pipeline.

Significance. If the performance advantage holds under more rigorous validation, the work would advance radar surveillance of small UAVs by retaining low-SNR signal information and enabling coherent multi-sensor fusion without intermediate detection losses. The explicit probabilistic treatment of unknown object cardinality and the derivation of a tractable VMP scheme from a joint model constitute clear technical contributions.

major comments (2)

[§4 (VMP derivation)] §4 (VMP derivation), mean-field factorization: the posterior is factorized as a product over per-object states, existences, and reflectivities. Because the likelihood is formed from the superimposed signal model (Eq. for the array response summed over objects), strong statistical dependence exists among nearby objects; the mean-field assumption therefore decouples precisely the interference terms that dominate the claimed operating regime. No error analysis or comparison against a less restrictive factorization is supplied, leaving the reported gains in closely-spaced simulations open to the possibility of optimistic bias.
[§5.2 (simulation results)] §5.2 (simulation results), performance comparison: the superiority is asserted for low-SNR and clutter-rich conditions, yet the manuscript provides neither the number of Monte-Carlo trials, confidence intervals on the reported metrics, nor a statistical test of the difference versus the super-resolution + BP baseline. Without these, it is impossible to judge whether the observed advantage is robust or an artifact of the particular synthetic data generator.

minor comments (2)

[§3 (probabilistic model)] The hierarchical Bernoulli-Gamma model is introduced without an explicit statement of its conjugacy properties or the resulting closed-form message expressions; adding these would improve reproducibility of the algorithm.
[Figures 3–5] Figure captions and axis labels should explicitly state the clutter density, RCS distribution, and sensor geometry used in each panel so that readers can map the plotted curves to the textual scenario descriptions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment point by point below, indicating planned revisions where appropriate.

read point-by-point responses

Referee: §4 (VMP derivation), mean-field factorization: the posterior is factorized as a product over per-object states, existences, and reflectivities. Because the likelihood is formed from the superimposed signal model (Eq. for the array response summed over objects), strong statistical dependence exists among nearby objects; the mean-field assumption therefore decouples precisely the interference terms that dominate the claimed operating regime. No error analysis or comparison against a less restrictive factorization is supplied, leaving the reported gains in closely-spaced simulations open to the possibility of optimistic bias.

Authors: We acknowledge the concern regarding the mean-field approximation. Although the variational posterior is factorized, the message updates are derived from the joint likelihood under the superimposed signal model, so the interference terms between closely spaced objects are explicitly incorporated via the expectation steps in the VMP iterations. This is a standard and tractable approach for such coupled models. To strengthen the presentation, we will add a dedicated paragraph in the revised Section 4 discussing the validity of the mean-field assumption in the low-SNR closely-spaced regime and include a limited comparison against a less restrictive (pairwise) factorization in a simplified two-object case to quantify approximation effects. revision: partial
Referee: §5.2 (simulation results), performance comparison: the superiority is asserted for low-SNR and clutter-rich conditions, yet the manuscript provides neither the number of Monte-Carlo trials, confidence intervals on the reported metrics, nor a statistical test of the difference versus the super-resolution + BP baseline. Without these, it is impossible to judge whether the observed advantage is robust or an artifact of the particular synthetic data generator.

Authors: We agree that additional statistical details are required to support the performance claims. In the revised Section 5.2, we will explicitly state the number of Monte Carlo trials used, include confidence intervals (or error bars) on all reported metrics such as OSPA distance and detection probability, and add results from appropriate statistical tests (e.g., paired t-tests) to assess the significance of improvements over the super-resolution + BP baseline. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation follows standard VMP from stated model

full rationale

The paper states a probabilistic model (superimposed signals + hierarchical Bernoulli-Gamma for existence/reflectivities), applies mean-field factorization, and derives message updates for VMP. This is a direct application of variational inference rules to the given joint posterior; the updates are not equivalent to the inputs by construction, nor are they obtained by fitting parameters to the target quantities or by self-citation chains. Simulation comparisons are empirical validation outside the derivation. No load-bearing self-citations, self-definitional steps, or renamed known results appear in the derivation chain.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 1 invented entities

The approach rests on these modeling choices and approximations, which are standard in probabilistic tracking but combined here in a new way.

free parameters (1)

Nuisance parameters
Learned directly from the radar signal as part of the algorithm.

axioms (2)

domain assumption The superimposed signal model accurately represents correlations from closely spaced objects.
Employed to capture data correlations in the raw sensor signals.
domain assumption Mean-field variational approximation is appropriate for this inference problem.
Used to derive computationally efficient message updates.

invented entities (1)

Hierarchical Bernoulli-Gamma model no independent evidence
purpose: Jointly models object existence, reflectivities, and reliability of radar-object links.
New model introduced to handle unknown and time-varying number of objects.

pith-pipeline@v0.9.0 · 5600 in / 1468 out tokens · 48368 ms · 2026-05-10T12:49:15.202629+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

49 extracted references · 1 canonical work pages

[1]

M. A. Richards,Fundamentals of Radar Signal Processing, 3rd ed. New York: McGraw-Hill Education, 2022

2022
[2]

Challenges in detecting UAS with radar,

P. Poitevin, M. Pelletier, and P. Lamontagne, “Challenges in detecting UAS with radar,” in2017 International Carnahan Conference on Secu- rity Technology (ICCST). IEEE, oct 2017

2017
[3]

Improved radar detection of small drones using doppler signal-to-clutter ratio (DSCR) detector,

J. Gong, J. Yan, H. Hu, D. Kong, and D. Li, “Improved radar detection of small drones using doppler signal-to-clutter ratio (DSCR) detector,” Drones, vol. 7, no. 5, p. 316, may 2023

2023
[4]

Drone detection and radar-cross-section measurements by RAD-DAR,

´A. D. Quevedo, F. I. Urzaiz, J. G. Menoyo, and A. A. L ´opez, “Drone detection and radar-cross-section measurements by RAD-DAR,”IET Radar, Sonar & Navigation, vol. 13, no. 9, pp. 1437–1447, Aug. 2019

2019
[5]

Review of the algorithms for radar single target tracking,

H. Wei, Z. ping Cai, B. Tang, and Z. xiang Yu, “Review of the algorithms for radar single target tracking,”IOP Conference Series: Earth and Environmental Science, vol. 69, p. 012073, jun 2017

2017
[6]

A CFAR adaptive matched filter detector,

F. Robey, D. Fuhrmann, E. Kelly, and R. Nitzberg, “A CFAR adaptive matched filter detector,”IEEE Trans. Aerosp. Electron. Syst., vol. 28, no. 1, pp. 208–216, Jan. 1992

1992
[7]

Bar-Shalom, P

Y . Bar-Shalom, P. K. Willett, and X. Tian,Tracking and Data Fusion: A Handbook of Algorithms. Storrs, CT, USA: Yaakov Bar-Shalom, 2011

2011
[8]

Mahler,Statistical Multisource-Multitarget Information Fusion

R. Mahler,Statistical Multisource-Multitarget Information Fusion. Nor- wood, MA, USA: Artech House, Inc., 2007

2007
[9]

Message passing algorithms for scalable multitarget tracking,

F. Meyer, T. Kropfreiter, J. L. Williams, R. Lau, F. Hlawatsch, P. Braca, and M. Z. Win, “Message passing algorithms for scalable multitarget tracking,”Proc. IEEE, vol. 106, no. 2, pp. 221–259, Feb. 2018

2018
[10]

Sequential detection and estimation of multipath channel parameters using belief propaga- tion,

X. Li, E. Leitinger, A. Venus, and F. Tufvesson, “Sequential detection and estimation of multipath channel parameters using belief propaga- tion,”IEEE Trans. Wireless Commun., vol. 21, no. 10, pp. 8385–8402, Apr. 2022

2022
[11]

Graph- based simultaneous localization and bias tracking,

A. Venus, E. Leitinger, S. Tertinek, F. Meyer, and K. Witrisal, “Graph- based simultaneous localization and bias tracking,”IEEE Trans. Wireless Commun., vol. 23, no. 10, pp. 13 141–13 158, May 2024

2024
[12]

Multisensor multiobject tracking with im- proved sampling efficiency,

W. Zhang and F. Meyer, “Multisensor multiobject tracking with im- proved sampling efficiency,”IEEE Transactions on Signal Processing, vol. 72, pp. 2036–2053, 2024

2036
[13]

Bernoulli track-before-detect filter for maritime radar,

B. Ristic, L. Rosenberg, D. Y . Kim, X. Wang, and J. Williams, “Bernoulli track-before-detect filter for maritime radar,”IET Radar, Sonar &; Navigation, vol. 14, no. 3, pp. 356–363, Feb. 2020

2020
[14]

Maximum likelihood track-before- detect with fluctuating target amplitude,

S. Tonissen and Y . Bar-Shalom, “Maximum likelihood track-before- detect with fluctuating target amplitude,”IEEE Trans. Aerosp. Electron. Syst., vol. 34, no. 3, pp. 796–809, Aug. 1998

1998
[15]

Recursive track-before-detect with target amplitude fluctuations,

M. Rutten, N. Gordon, and S. Maskell, “Recursive track-before-detect with target amplitude fluctuations,”IEE Proceedings - Radar, Sonar and Navigation, vol. 152, no. 5, p. 345, 2005

2005
[16]

A comparison of detec- tion performance for several track-before-detect algorithms,

S. J. Davey, M. G. Rutten, and B. Cheung, “A comparison of detec- tion performance for several track-before-detect algorithms,”EURASIP Journal on Advances in Signal Processing, vol. 2008, no. 1, nov 2007

2008
[17]

A multi-dimensional hough transform-based track-before-detect technique for detecting weak targets in strong clutter backgrounds,

L. R. Moyer, J. Spak, and P. Lamanna, “A multi-dimensional hough transform-based track-before-detect technique for detecting weak targets in strong clutter backgrounds,”IEEE Trans. Aerosp. Electron. Syst., vol. 47, no. 4, pp. 3062–3068, Oct. 2011

2011
[18]

Generalized labeled multi-Bernoulli approximation of multi-object densities,

F. Papi, B.-N. V o, B.-T. V o, C. Fantacci, and M. Beard, “Generalized labeled multi-Bernoulli approximation of multi-object densities,”IEEE Trans. Aerosp. Electron. Syst., vol. 63, no. 20, pp. 5487–5497, July 2015

2015
[19]

Multitarget likelihood com- putation for track-before-detect applications with amplitude fluctuations of type Swerling 0, 1, and 3,

A. Lepoutre, O. Rabaste, and F. Le Gland, “Multitarget likelihood com- putation for track-before-detect applications with amplitude fluctuations of type Swerling 0, 1, and 3,”IEEE Trans. Aerosp. Electron. Syst., vol. 52, no. 3, pp. 1089–1107, June 2016

2016
[20]

Recursive Bayesian filtering for multitarget track-before- detect in passive radars,

F. Lehmann, “Recursive Bayesian filtering for multitarget track-before- detect in passive radars,”IEEE Trans. Aerosp. Electron. Syst., vol. 48, no. 3, pp. 2458–2480, Jul. 2012

2012
[21]

Multiple model efficient particle filter based track-before-detect for maneuvering weak targets,

B. Zhichao, J. Qiuxi, and L. Fangzheng, “Multiple model efficient particle filter based track-before-detect for maneuvering weak targets,” Journal of Systems Engineering and Electronics, vol. 31, no. 4, pp. 647– 656, aug 2020

2020
[22]

Dynamic programming solution for detecting dim moving targets,

Y . Barniv, “Dynamic programming solution for detecting dim moving targets,”IEEE Trans. Aerosp. Electron. Syst., vol. AES-21, no. 1, pp. 144–156, Feb. 1985

1985
[23]

Multiobject tracking for thresholded cell measurements,

T. Kropfreiter, J. L. Williams, and F. Meyer, “Multiobject tracking for thresholded cell measurements,” in2024 27th International Conference on Information Fusion (FUSION). IEEE, Jul. 2024, pp. 1–8

2024
[24]

A bernoulli track- before-detect filter for interacting targets in maritime radar,

D. Y . Kim, B. Ristic, R. Guan, and L. Rosenberg, “A bernoulli track- before-detect filter for interacting targets in maritime radar,”IEEE Trans. Aerosp. Electron. Syst., vol. 57, no. 3, pp. 1981–1991, Jan. 2021

1981
[25]

A BP method for track-before- detect,

M. Liang, T. Kropfreiter, and F. Meyer, “A BP method for track-before- detect,”IEEE Signal Processing Letters, vol. 30, pp. 1137–1141, 2023

2023
[26]

A scalable track-before- detect method with Poisson/multi-Bernoulli model,

T. Kropfreiter, J. L. Williams, and F. Meyer, “A scalable track-before- detect method with Poisson/multi-Bernoulli model,” inProc. FUSION- 2021, Dec. 2021, pp. 1–8

2021
[27]

A belief propagation approach for direct multipath-based SLAM,

M. Liang, E. Leitinger, and F. Meyer, “A belief propagation approach for direct multipath-based SLAM,” inProc. Asilomar-23, Pacific Grove, CA, USA, Nov. 2023

2023
[28]

Direct multipath-based SLAM,

——, “Direct multipath-based SLAM,”IEEE Trans. Signal Process., vol. 73, pp. 2336–2352, Mar. 2025

2025
[29]

An approach of directly tracking multiple objects,

M. Liang and F. Meyer, “An approach of directly tracking multiple objects,” inProc. Asilomar-24, Oct. 2024, pp. 995–999

2024
[30]

Message passing for track-before-detect,

——, “Message passing for track-before-detect,”ArXiv e-prints, 2025. [Online]. Available: https://arxiv.org/abs/2506.18177

work page arXiv 2025
[31]

Maximum-likelihood direct position estimation in dense multipath,

O. Bialer, D. Raphaeli, and A. Weiss, “Maximum-likelihood direct position estimation in dense multipath,”IEEE Trans. Veh. Technol., vol. 62, no. 5, pp. 2069–2079, Jun. 2013

2069
[32]

Variational Bayesian expectation maximization for radar map estimation,

M. Lundgren, L. Svensson, and L. Hammarstrand, “Variational Bayesian expectation maximization for radar map estimation,”IEEE Trans. Signal Process., vol. 64, no. 6, pp. 1391–1404, Mar. 2016

2016
[33]

Variational tracking and redetection for closely-spaced objects in heavy clutter,

R. Gan, Q. Li, and S. J. Godsill, “Variational tracking and redetection for closely-spaced objects in heavy clutter,”IEEE Trans. Aerosp. Electron. Syst., vol. 60, no. 4, pp. 5286–5311, May 2024

2024
[34]

Robust multitarget tracking in interference environments: A message-passing approach,

X. Bai, H. Lan, Z. Wang, Q. Pan, Y . Hao, and C. Li, “Robust multitarget tracking in interference environments: A message-passing approach,” IEEE Trans. Aerosp. Electron. Syst., vol. 60, no. 1, pp. 360–386, Oct. 2024

2024
[35]

Histogram-PMHT unfettered,

S. J. Davey, M. Wieneke, and H. Vu, “Histogram-PMHT unfettered,” IEEE J. Sel. Topics Signal Process., vol. 7, no. 3, pp. 435–447, Mar. 2013

2013
[36]

Bayesian joint localization and tracking algorithm using multiple-input multiple-output radar,

A. M. Westerkam, C. N. Manch ´on, P. Mogensen, and T. Pedersen, “Bayesian joint localization and tracking algorithm using multiple-input multiple-output radar,” inProc. IEEE CAMSAP-23, Dec. 2023

2023
[37]

Variational Bayesian inference of line spectra,

M.-A. Badiu, T. L. Hansen, and B. H. Fleury, “Variational Bayesian inference of line spectra,”IEEE Transactions on Signal Processing, vol. 65, no. 9, pp. 2247–2261, May 2017

2017
[38]

Variational inference of structured line spectra exploiting group-sparsity,

J. M ¨oderl, E. Leitinger, F. Pernkopf, and K. Witrisal, “Variational inference of structured line spectra exploiting group-sparsity,”IEEE Trans. Signal Process., Nov. 2024

2024
[39]

A sparse Bayesian learning algorithm with dictionary parameter estimation,

T. L. Hansen, M. A. Badiu, B. H. Fleury, and B. D. Rao, “A sparse Bayesian learning algorithm with dictionary parameter estimation,” in Proc. IEEE SAM 2014, Jun. 2014, pp. 385–388

2014
[40]

Super-resolution estimation of UWB channels including the dense component – An SBL- inspired approach,

S. Grebien, E. Leitinger, K. Witrisal, and B. H. Fleury, “Super-resolution estimation of UWB channels including the dense component – An SBL- inspired approach,”IEEE Trans. Wireless Commun., vol. 23, no. 8, pp. 10 301–10 318, Feb. 2024

2024
[41]

A block-sparse Bayesian learning algorithm with dictionary parameter estimation for multi-sensor data fusion,

J. M ¨oderl, A. M. Westerkam, A. Venus, and E. Leitinger, “A block-sparse Bayesian learning algorithm with dictionary parameter estimation for multi-sensor data fusion,” inProc. Fusion-2025, Brasil, Rio De Janeiro, Jul. 2025

2025
[42]

Variational message passing-based multiobject tracking for MIMO-radars using raw sensor signals,

A. M. Westerkam, J. M ¨oderl, E. Leitinger, and T. Pedersen, “Variational message passing-based multiobject tracking for MIMO-radars using raw sensor signals,” inProc. Fusion-2025, Rio de Janeiro, Brazil, Mar. 2025

2025
[43]

C. M. Bishop,Pattern Recognition and Machine Learning (Information Science and Statistics), 1st ed. Springer, 2007

2007
[44]

Distributed algorithm for cooperative joint localization and tracking using multiple-inputmultiple-output radars,

A. H. F. Kitchen, M. S. L. Brøndt, M. S. Jensen, T. Pedersen, and A. M. Westerkam, “Distributed algorithm for cooperative joint localization and tracking using multiple-inputmultiple-output radars,” inProc. IEEE RadarCon-25, Atlanta, USA, May 2025, p. 6

2025
[45]

A consistent metric for performance evaluation of multi-object filters,

D. Schuhmacher, B.-T. V o, and B.-N. V o, “A consistent metric for performance evaluation of multi-object filters,”IEEE Transactions on Signal Processing, vol. 56, no. 8, pp. 3447–3457, Aug. 2008

2008
[46]

Neural expectation maximization,

K. Greff, S. van Steenkiste, and J. Schmidhuber, “Neural expectation maximization,” inProc. NeuRIPS-2017, Red Hook, NY , USA, Dec. 2017, pp. 6694–6704

2017
[47]

A neural-enhanced factor graph-based algorithm for robust positioning in obstructed LOS situations,

A. Venus, E. Leitinger, S. Tertinek, and K. Witrisal, “A neural-enhanced factor graph-based algorithm for robust positioning in obstructed LOS situations,”IEEE Open J. of Signal Process., vol. 5, pp. 29–38, Nov. 2023

2023
[48]

Bayesian multiobject tracking with neural-enhanced motion and measurement models,

S. Wei, M. Liang, and F. Meyer, “Bayesian multiobject tracking with neural-enhanced motion and measurement models,”IEEE Trans. Signal Process., pp. 1–16, Jan. 2026

2026
[49]

Kullback-leibler divergence and moment matching for hyperspherical probability distributions,

G. Kurz, F. Pfaff, and U. D. Hanebeck, “Kullback-leibler divergence and moment matching for hyperspherical probability distributions,” inProc. FUSION 2016, 2016, pp. 2087–2094

2016