Pith. sign in

REVIEW 3 major objections 4 minor 54 references

A training-free variational tracker rivals deep-learning radar detection, running in real time on raw point clouds.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

PiVoT uses a two-stage variational inference scheme to jointly detect, track, and shape-estimate up to about 1000 objects in heavy clutter, in real time and without training.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection PiVoT is a real algorithmic contribution, but its headline clutter-accuracy relies on an approximation validated only for isolated objects; fix that and this is a strong paper. the 3 major comments →

arxiv 2607.13891 v1 pith:AHW2K5C7 submitted 2026-07-15 cs.LG cs.CVeess.SPstat.ML

PiVoT: A Variational Solution for Real-time Large-scale Multi-object Detection and Tracking under Heavy Clutter

Runze Gan , Qing Li , Simon J. Godsill , Mike E. Davies , James R. Hopgood This is my paper
classification cs.LG cs.CVeess.SPstat.ML MSC 60G5562F1562M2068T37
keywords multi-object trackingpoint cloudsvariational inferencePoisson measurement modelDoppler radardetectionexistence probabilitiesclutter
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

PiVoT is a training-free, model-based multi-object tracker that works directly on noisy radar point clouds, jointly estimating object kinematics, shapes, existence probabilities, measurement rates, and data association. The paper claims that a two-stage variational inference scheme—coordinate-ascent updates in Stage 1, and a globally optimal linear-time existence evaluation in Stage 2—lets this approach outperform existing Bayesian trackers in heavy clutter and scale to a thousand objects, while running in real time on the full-scale RadarScenes automotive radar dataset at performance comparable to a deep-learning detection benchmark. A Doppler-augmented Poisson measurement model preserves linear-Gaussian updates, so Doppler velocities are exploited without pruning near-zero returns. A sympathetic reader would care because this offers a route to radar perception in data-scarce settings where labelled training data are expensive or unavailable.

Core claim

The central claim is that the intractable joint filtering posterior over states, rates, shapes, associations, existence, and detectability can be approximated well by factorising it into two stages: Stage 1 marginalises out existence and detectability and applies coordinate-ascent variational inference for kinematics, rate, shape, and association; Stage 2, with those factors fixed, computes the globally optimal existence/detectability distribution for every object in O(K_n M) time via a closed-form expression. The key approximation enabling closed forms is replacing each object's random detectability by its mean in the association prior, justified as the KLD-optimal NHPP approximation and an

What carries the argument

Two-stage variational inference built on a Poisson measurement model. Stage 1 uses closed-form coordinate-ascent updates for Gaussian states, Gamma rates, Wishart shapes, and categorical associations, with a theorem-backed pruning rule (Lambert-W threshold) that removes births whose total association mass falls below a computable bound. Stage 2 derives the globally optimal existence/detectability posterior for each object in a form that collapses the dependence on association, reducing complexity from O(K^2 M) to O(K M). The load-bearing approximation (22) replaces each object's detectability indicator D_k by its mean p_e' p_d in the association term, turning the marginalised process into an

Load-bearing premise

The paper's main approximation is Eq. (22), which replaces each object's random detectability indicator by its mean when computing association and existence updates; the accuracy guarantee is only demonstrated for objects well separated from others, while the headline results involve dense clutter and coalescing objects.

What would settle it

Simulate a small scene (two or three closely spaced objects, detection probability 0.9, heavy clutter) and enumerate all measurement-to-object associations to obtain exact existence and association posteriors. Compare with PiVoT's Stage-2 probabilities; if systematic deviations appear—e.g., existence probability off by more than 0.05, or repeated failures to detect one object in coalescence—then approximation (22), not the inference machinery, is the source, and the dense-scene accuracy claim is undercut.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Real-time, training-free detection and tracking on full-resolution automotive radar point clouds becomes practical, with processing below the 1-second data interval on a laptop CPU.
  • Surveillance and drone-swarm tracking can scale to roughly a thousand objects without gating, completing a time step in under a second on a laptop.
  • Object existence and detectability probabilities become part of the tracking output, giving users uncertainty estimates about whether a tracked object really exists, at negligible extra cost.
  • Near-zero Doppler returns, often pruned as stationary clutter, are exploited as evidence for transversely moving objects, improving recall in automotive scenes.
  • Joint detection and tracking no longer requires external clustering or a separate detector; clustering emerges from the variational association updates.

Where Pith is reading between the lines

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

  • The linear-time existence formula (37)–(38) is a portable algorithmic piece: any NHPP-style extended-object tracker could reuse the same shared-sums trick to evaluate object existences without quadratic association enumeration.
  • The accuracy guarantee for the core approximation covers only well-separated objects, so the method's headline regime—dense, coalescing scenes—is exactly where approximation bias is least protected; a targeted exact-marginalization study on small dense scenes would map where it starts to matter.
  • If the central claim holds, the deterministic report-stage heuristics (ghost suppression, Doppler-informativeness gates) could in principle be folded back into the probabilistic model, so that the final moving-object declaration comes directly from the posterior.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. PiVoT is a training-free Bayesian multi-object tracker for point-cloud measurements under an NHPP extended-object model with uncertain existence and detectability. It uses a two-stage variational inference framework: Stage 1 runs closed-form coordinate-ascent updates for kinematic states, Poisson rates, shape precision matrices, and data association; Stage 2 evaluates per-object existence/detectability in O(K_n M) time. The paper contributes a Doppler-augmented NHPP model with a linear-Gaussian per-object likelihood, a birth-pruning procedure justified by Theorem IX.1, and numerical experiments on six simulated datasets, a 1000-object demo, and the RadarScenes automotive radar benchmark, where PiVoT is compared with PMBM, SPA, and the deep-learning RadarGNN baseline.

Significance. If the reported results hold, PiVoT would be a substantial advance: a fully training-free joint detector/tracker that scales to hundreds or thousands of objects, runs in real time on full-resolution automotive radar point clouds, and is competitive with a deep-learning detection benchmark. The main technical strengths are the closed-form CAVI updates, the O(K_n M) Stage-2 existence evaluation, the Doppler model that preserves Gaussian conjugacy, and the explicit theoretical analysis of birth pruning. These contributions are meaningful beyond the particular radar application and could benefit other NHPP-based extended-object tracking problems. The central caveat is that nearly all closed-form expressions depend on the mean-field detectability approximation in Eq. (22), whose error is analyzed only for isolated objects; the headline dense-clutter and coalescence regimes are precisely where that analysis does not apply.

major comments (3)
  1. [Section III-C, Eq. (22); Appendix D] The approximation in Eq. (22), replacing D_k by p'_e,k p_d,k in the association term Xi_k, is used in the Stage 1 ELBO (23), all CAVI updates (27)-(32), and the Stage 2 existence formula (37)-(40). It is the only approximation that separates PiVoT from the exact NHPP model. Appendix C justifies the approximation as KLD-optimal only within the class of NHPPs, but the exact marginalized process is a mixture of an empty process and an NHPP; the KLD argument does not control errors in association weights or existence probabilities. Appendix D explicitly considers an object well separated from all others, where gating removes cross-object contributions. The paper's headline regimes—DS4-DS6 heavy clutter, the 1000-object coalescence demo, and RadarScenes—are precisely the regimes where this separation fails. The optional Stage 2 refinement (Appendix D-2) is described but not run in any reporte
  2. [Section IV-C, Section IX, Appendix J] The implemented birth-pruning threshold is a heuristic L=0.5 (or L=0.05 as an alternative), while Theorem IX.1 defines a sufficient threshold L_k(s) under Assumption 1. These are not the same object: the theorem guarantees monotone decay of association mass for a removed birth under specific modelling conditions, whereas the experiments use a fixed scalar threshold with no certified link to the theorem. Since pruning removes births before convergence, an overly large L can discard valid detections; the reported accuracy results may depend on this choice. Please report results for several L values (e.g., 0.05, 0.5, and the certified L_k(s) when computable), and state whether Assumption 1 holds in each experiment, especially the uniform-clutter-intensity condition on RadarScenes.
  3. [Section VII-C, Appendix G-2, Table II] The RadarScenes moving-object output is produced by a hand-specified deterministic post-processing cascade (speed gates, Doppler-informativeness checks, concrete-shape plausibility, ghost suppression). This is not learned, but it is also not part of the variational NHPP inference. The paper's claim of a 'fully training-free joint detector and tracker' with performance comparable to a deep-learning benchmark applies to this post-processed output. The precision/F1 gains in Table II relative to RadarGNN could arise substantially from these manually tuned criteria rather than from the underlying inference. Please ablate the post-processing stages (raw tracks, after each check) and clarify how much of the reported F1/GOSPA improvement is attributable to the variational tracker itself versus the moving-object selection stage.
minor comments (4)
  1. [Section V] The phrase 'globally optimal' for q*_2(E_k,D_k) is correct only with respect to the approximate Stage 2 ELBO F^k_2, which itself uses the Eq. (22) approximation. The paper is transparent about this, but the abstract and contributions should avoid implying exact-model global optimality.
  2. [Section VIII-C, Table II] The RadarScenes comparison uses different frame accumulation lengths (0.25 s for PiVoT vs 0.5 s for RadarGNN) and different input features (PiVoT omits RCS). The paper acknowledges this, but the comparison would be strengthened by reporting RadarGNN with the shorter accumulation and PiVoT with RCS as an additional input, if feasible.
  3. [Table I] The table formatting is difficult to parse as printed; consider splitting DS1-DS6 into separate sub-tables or using clearer column grouping. Also note in the caption that SPA15000 is omitted on DS3 and SPA100 uses only 10 runs on DS4/DS6.
  4. [Section IV-C, Remark 3] The paper states that L=0.5 is used 'for speed' and L=0.05 'for reliability', but no guidance is given for choosing L in new scenes. A brief sensitivity plot of GOSPA vs L on DS4 or DS5 would help practitioners.

Circularity Check

0 steps flagged

No significant circularity: PiVoT's derivation is self-contained and its headline claims are benchmarked externally.

full rationale

The paper's derivation chain does not reduce to its inputs. Stage 1 (Eq. 23) is built from the exact NHPP likelihood (4)-(6) plus the explicitly stated approximation (22); that approximation is justified as a first-order Taylor expansion and as the KLD-optimal NHPP projection (Appendix C, Lemma C.1), not by fitting a parameter to the target existence or association outputs. All CAVI updates (26)-(32) follow from the standard conditional-update formula (24), and the Stage 2 'globally optimal' existence formula (37)-(40) is derived in Appendix E as the exact optimum of the stated approximate Stage 2 ELBO, not assumed to equal the exact posterior. The birth-pruning result (Theorem IX.1) is proved in Appendices H-I under Assumption 1, rather than imported by self-citation. The performance claims are measured against external baselines (PMBM, SPA, RadarGNN) and ground-truth GOSPA/F1 metrics; no headline number is a refit of the same quantity used to define the method. Several citations are to the authors' own prior CAVI/NHPP work, but they are contextual; the one design heuristic attributed to [22] (Section IV-D) is also empirically demonstrated in Fig. 3, so the self-citations are not load-bearing. The main residual concern is the accuracy of approximation (22) in dense/clutter regimes—a correctness/robustness risk, not a circularity.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 0 invented entities

The central claim rests on the NHPP measurement model, a paper-specific detectability-averaging approximation, and a substantial set of hand-set hyperparameters and post-processing rules. The variational math is self-contained, but the empirical headline results depend on choices that are not inferred from data and are partly tuned on the validation split.

free parameters (7)
  • birth pruning threshold L = 0.5 (speed) or 0.05 (conservative)
    Chosen heuristically; Theorem IX.1 gives a sufficient threshold L_k(s), but the implementation does not compute or use it. Affects detection reliability and speed.
  • existence/reporting thresholds P_pru, P_rep, P_stp = 0.03, 0.8, 0.3 in simulations
    Hand-set; they control which tracks are pruned, reported, and kept being reported, directly affecting GOSPA and precision/recall.
  • moving-object post-processing criteria = not numerically specified
    Minimum speed gates, velocity Mahalanobis thresholds, window lengths, shape constraints, and ghost checks are selected by visual inspection of validation videos; they materially affect the RadarScenes comparison.
  • clutter rate lambda_0 = current number of measurements M_n
    PiVoT treats the unknown clutter rate as a crude upper bound equal to the number of measurements rather than estimating it (Appendix J).
  • Doppler noise variance sigma_v^2 = not stated in main text
    Required in the Doppler likelihood (41) and association updates (45), but no value or sensitivity analysis is given.
  • birth/survival/forgetting hyperparameters p_b, p_s, gamma_Lambda, gamma_P = 0.2, 0.9, 0.9, 0.9 in simulations
    Hand-set in Appendix J; they control the birth count, survival, and rate/shape prediction uncertainty.
  • birth tiling grid size and placement = 1312 initial births in Fig. 3
    Tile size and spacing determine detection resolution and initial K_b; no automatic rule is given.
axioms (6)
  • domain assumption NHPP extended-object measurement model with Poisson cardinality and spatial density
    Taken from [11] and used as the foundational measurement model throughout Section II-A.
  • domain assumption Conjugate Gaussian/Gamma/Wishart prior and transition structure
    Assumed in Section II-B and used to obtain closed-form CAVI updates; departures from these families would break the method's efficiency.
  • domain assumption Detection points from one object share the same velocity
    Stated in Section VI-A; valid for rigid-body translation but not turning or micro-Doppler, as the paper acknowledges.
  • ad hoc to paper Approximation Eq. (22): marginalised detectability replaced by an NHPP of mean rate
    All Stage 1 and Stage 2 closed-form updates depend on this replacement. It is justified as KLD-minimizing in Appendix C, but only analysed for isolated objects in Appendix D.
  • domain assumption Assumption 1 for the birth-pruning theorem: uniform clutter and flat uninformative birth prior
    Assumption 1 in Section IX is required by Theorem IX.1; it holds in the simulated setting but not on real RadarScenes clutter.
  • domain assumption Survival/existence/detectability conditional independence across objects and features
    Used in Sections II-B and II-C to factorise priors and posteriors; without it the variational factorisation and updates change.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PiVoT: A Variational Solution for Real-time Large-scale Multi-object Detection and Tracking under Heavy Clutter." pith.science (2026). https://pith.science/paper/AHW2K5C7

@misc{pith2026260713891,
  author       = {Pith},
  title        = {Pith review of: PiVoT: A Variational Solution for Real-time Large-scale Multi-object Detection and Tracking under Heavy Clutter},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AHW2K5C7}},
  note         = {Machine review of arXiv:2607.13891}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Multi-object detection and tracking from noisy point clouds remain challenging in many data-scarce radar applications. Current Bayesian trackers based on Poisson measurement models offer a training-free solution but struggle to achieve accuracy and efficiency under severe clutter, large object populations, and full-resolution Doppler point clouds. We address this with PiVoT, a fast, clutter-resilient multi-object tracker for both positional and Doppler measurements. PiVoT performs end-to-end detection and tracking of a large and time-varying number of objects without external clustering or detectors, through joint inference of object states, shapes, existence probabilities, data association, and measurement rates. Its efficiency is driven by several variational inference innovations, such as theoretically justified birth pruning, quadratic-to-linear complexity reductions for exact updates, and a computationally efficient Doppler Poisson model. Experiments show that PiVoT substantially outperforms existing Bayesian trackers in challenging scenes, while also demonstrating exceptional scalability to a thousand objects, robustness to clutter visually inseparable from objects, and real-time operation on full-scale modern automotive radar datasets, where it attains performance comparable to a deep-learning detection benchmark as a training-free joint detector and tracker.

Figures

Figures reproduced from arXiv: 2607.13891 by James R. Hopgood, Mike E. Davies, Qing Li, Runze Gan, Simon J. Godsill.

Figure 1
Figure 1. Figure 1: PiVoT joint detection and tracking on real radar data, with associated workflow. (a) Camera image from the RadarScenes [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Simulated NHPP measurements from three objects over [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Clutter-robust detection using PiVoT Stage 1. The data is from Dataset 3 (first time step) in Section VIII-A, showing [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) Automotive radar data from [8], with PiVoT estimates of moving objects overlaid (ellipses: shape, curves: tracks). [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: PiVoT tracking results on the RadarScenes [8] dataset [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: True tracks (black), shapes, one-run PiVoT estimates (red dashed tracks and ellipses), and final-time-step measurements [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The challenge of this case is efficiently handling a [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Tracking results for up to 1036 objects. Black curves [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualisation on representative RadarScenes frames. The results are matched to the camera images. Coloured points [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references

  1. [1]

    Sequential point clouds: A survey,

    H. Wang and Y . Tian, “Sequential point clouds: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 8, pp. 5504–5523, 2024

    2024

  2. [2]

    A tutorial on multiple extended object tracking,

    K. Granstr ¨om and M. Baum, “A tutorial on multiple extended object tracking,”Authorea Preprints, 2022

    2022

  3. [3]

    Object detection for automotive radar point clouds–a comparison,

    N. Scheiner, F. Kraus, N. Appenrodt, J. Dickmann, and B. Sick, “Object detection for automotive radar point clouds–a comparison,”AI Perspectives, vol. 3, no. 1, p. 6, 2021

    2021

  4. [4]

    SpaRTA tracking across occlusions via partitioning of 3D clouds of points,

    A. Cavagna, S. Melillo, L. Parisi, and F. Ricci-Tersenghi, “SpaRTA tracking across occlusions via partitioning of 3D clouds of points,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 4, pp. 1394–1403, 2019

    2019

  5. [5]

    Evolutionary developments of to- day’s remote sensing radar technology—right from the telemobiloscope: A review,

    S. S. Kari, A. A. B. Raj, and B. K, “Evolutionary developments of to- day’s remote sensing radar technology—right from the telemobiloscope: A review,”IEEE Geoscience and Remote Sensing Magazine, vol. 12, no. 1, pp. 67–107, 2023

    2023

  6. [6]

    A review of automatic classification of drones using radar: key considerations, performance evaluation, and prospects,

    B. I. Ahmad, C. Rogers, S. Harman, H. Dale, M. Jahangir, M. Antoniou, C. Baker, M. Newman, and F. Fioranelli, “A review of automatic classification of drones using radar: key considerations, performance evaluation, and prospects,”IEEE Aerospace and Electronic Systems Magazine, vol. 39, no. 2, pp. 18–33, 2023

    2023

  7. [7]

    New chal- lenges for deep neural networks in automotive radar perception: An overview of current research trends,

    N. Scheiner, F. Weishaupt, J. F. Tilly, and J. Dickmann, “New chal- lenges for deep neural networks in automotive radar perception: An overview of current research trends,”Automatisiertes Fahren 2020: Von der Fahrerassistenz zum autonomen Fahren 6. Internationale ATZ- Fachtagung, pp. 165–182, 2021

    2020

  8. [8]

    RadarScenes: A real-world radar point cloud data set for automotive applications,

    O. Schumann, M. Hahn, N. Scheiner, F. Weishaupt, J. F. Tilly, J. Dick- mann, and C. W ¨ohler, “RadarScenes: A real-world radar point cloud data set for automotive applications,” in2021 IEEE 24th International Conference on Information Fusion (FUSION). IEEE, 2021, pp. 1–8

    2021

  9. [9]

    Bar-Shalom and X.-R

    Y . Bar-Shalom and X.-R. Li,Multitarget-multisensor tracking: princi- ples and techniques. YBs Storrs, CT, 1995, vol. 19

    1995

  10. [10]

    An efficient implementation of Reid’s multiple hypothesis tracking algorithm and its evaluation for the purpose of visual tracking,

    I. J. Cox and S. L. Hingorani, “An efficient implementation of Reid’s multiple hypothesis tracking algorithm and its evaluation for the purpose of visual tracking,”IEEE Transactions on pattern analysis and machine intelligence, vol. 18, no. 2, pp. 138–150, 1996

    1996

  11. [11]

    Poisson models for extended target and group tracking,

    K. Gilholm, S. Godsill, S. Maskell, and D. Salmond, “Poisson models for extended target and group tracking,” inSignal and Data Processing of Small Targets 2005, vol. 5913. SPIE, 2005, pp. 230–241

    2005

  12. [12]

    Poisson multi-Bernoulli mixture conjugate prior for multiple extended target filtering,

    K. Granstr ¨om, M. Fatemi, and L. Svensson, “Poisson multi-Bernoulli mixture conjugate prior for multiple extended target filtering,”IEEE Transactions on Aerospace and Electronic Systems, vol. 56, no. 1, pp. 208–225, 2019

    2019

  13. [13]

    Trajectory PMB filters for extended object tracking using belief propagation,

    Y . Xia, ´A. F. Garc´ıa-Fern´andez, F. Meyer, J. L. Williams, K. Granstr ¨om, and L. Svensson, “Trajectory PMB filters for extended object tracking using belief propagation,”IEEE Transactions on Aerospace and Elec- tronic Systems, vol. 59, no. 6, pp. 9312–9331, 2023

    2023

  14. [14]

    Scalable detection and tracking of geometric extended objects,

    F. Meyer and J. L. Williams, “Scalable detection and tracking of geometric extended objects,”IEEE Transactions on Signal Processing, vol. 69, pp. 6283–6298, 2021. 17

    2021

  15. [15]

    A scalable Rao-Blackwellised sequential MCMC sampler for joint detection and tracking in clutter,

    Q. Li, R. Gan, and S. Godsill, “A scalable Rao-Blackwellised sequential MCMC sampler for joint detection and tracking in clutter,” in2023 26th International Conference on Information Fusion (FUSION). IEEE

  16. [16]

    An adaptive and scalable multi-object tracker based on the non-homogeneous Poisson process,

    Q. Li, R. Gan, J. Liang, and S. J. Godsill, “An adaptive and scalable multi-object tracker based on the non-homogeneous Poisson process,” IEEE Transactions on Signal Processing, vol. 71, pp. 105–120, 2023

    2023

  17. [17]

    Chapter 3 - target detection,

    A. Bole, A. Wall, and A. Norris, “Chapter 3 - target detection,” inRadar and ARPA Manual, 3rd ed. Butterworth-Heinemann, 2014, pp. 139–213

    2014

  18. [18]

    PiV oT project page and demos: https://runzegan.github.io/projects/pivot/

  19. [19]

    Variational inference: A review for statisticians,

    D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, “Variational inference: A review for statisticians,”Journal of the American statistical Associa- tion, vol. 112, no. 518, pp. 859–877, 2017

    2017

  20. [20]

    Bishop,Pattern Recognition and Machine Learning

    C. Bishop,Pattern Recognition and Machine Learning. Springer, 2006

    2006

  21. [21]

    A variational Bayes association-based multi-object tracker under the non-homogeneous Poisson measurement process,

    R. Gan, Q. Li, and S. Godsill, “A variational Bayes association-based multi-object tracker under the non-homogeneous Poisson measurement process,” in2022 25th International Conference on Information Fusion (FUSION). IEEE, 2022, pp. 1–8

    2022

  22. [22]

    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 Transactions on Aerospace and Electronic Systems, vol. 60, no. 4, pp. 5286–5311, 2024

    2024

  23. [23]

    A complete variational tracker,

    R. D. Turner, S. Bottone, and B. Avasarala, “A complete variational tracker,”Advances in Neural Information Processing Systems, 2014

    2014

  24. [24]

    A structured mean field approach for existence-based multiple target tracking,

    R. A. Lau and J. L. Williams, “A structured mean field approach for existence-based multiple target tracking,” in2016 19th International Conference on Information Fusion (FUSION). IEEE, 2016

    2016

  25. [25]

    Integrated track maintenance for the PMHT via the hysteresis model,

    S. J. Davey and D. A. Gray, “Integrated track maintenance for the PMHT via the hysteresis model,”IEEE transactions on Aerospace and Electronic Systems, vol. 43, no. 1, pp. 93–111, 2007

    2007

  26. [26]

    A direct scattering model for tracking vehicles with high-resolution radars,

    C. Knill, A. Scheel, and K. Dietmayer, “A direct scattering model for tracking vehicles with high-resolution radars,” in2016 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2016, pp. 298–303

    2016

  27. [27]

    Incorporating range rate measurements in ekf-based elliptical extended object tracking,

    K. Thormann and M. Baum, “Incorporating range rate measurements in ekf-based elliptical extended object tracking,” in2021 IEEE Interna- tional Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI). IEEE, 2021, pp. 1–6

    2021

  28. [28]

    Tracking multiple vehicles using a vari- ational radar model,

    A. Scheel and K. Dietmayer, “Tracking multiple vehicles using a vari- ational radar model,”IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 10, pp. 3721–3736, 2018

    2018

  29. [29]

    Detection and tracking on automotive radar data with deep learning,

    J. F. Tilly, S. Haag, O. Schumann, F. Weishaupt, B. Duraisamy, J. Dick- mann, and M. Fritzsche, “Detection and tracking on automotive radar data with deep learning,” in2020 IEEE 23rd International Conference on Information Fusion (FUSION). IEEE, 2020, pp. 1–7

    2020

  30. [30]

    A multi- stage clustering framework for automotive radar data,

    N. Scheiner, N. Appenrodt, J. Dickmann, and B. Sick, “A multi- stage clustering framework for automotive radar data,” in2019 IEEE Intelligent Transportation Systems Conference (ITSC). IEEE, 2019, pp. 2060–2067

    2019

  31. [31]

    Constraint-based hierarchical cluster selection in automotive radar data,

    C. Malzer and M. Baum, “Constraint-based hierarchical cluster selection in automotive radar data,”Sensors, vol. 21, no. 10, p. 3410, 2021

    2021

  32. [32]

    PiV oT: Poisson measurements-based variational multi-object detection and tracking,

    R. Gan, Q. Li, J. R. Hopgood, M. E. Davies, and S. Godsill, “PiV oT: Poisson measurements-based variational multi-object detection and tracking,” in2025 28th International Conference on Information Fusion (FUSION). IEEE, 2025, pp. 1–8

    2025

  33. [33]

    Papoulis and S

    A. Papoulis and S. U. Pillai,Probability, Random Variables, and Stochastic Processes, 4th ed. McGraw-Hill, 2002, Section 5-4

    2002

  34. [34]

    Conditionally factorized variational Bayes with importance sampling,

    R. Gan and S. J. Godsill, “Conditionally factorized variational Bayes with importance sampling,” inIEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP). IEEE, 2022

    2022

  35. [35]

    Bayesian approach to extended object and cluster tracking using random matrices,

    J. W. Koch, “Bayesian approach to extended object and cluster tracking using random matrices,”IEEE Transactions on Aerospace and Electronic Systems, vol. 44, no. 3, pp. 1042–1059, 2008

    2008

  36. [36]

    General- ized optimal sub-pattern assignment metric,

    A. S. Rahmathullah, ´A. F. Garc´ıa-Fern´andez, and L. Svensson, “General- ized optimal sub-pattern assignment metric,” in2017 20th International Conference on Information Fusion (FUSION). IEEE, 2017, pp. 1–8

    2017

  37. [37]

    RadarGNN: Transfor- mation invariant graph neural network for radar-based perception,

    F. Fent, P. Bauerschmidt, and M. Lienkamp, “RadarGNN: Transfor- mation invariant graph neural network for radar-based perception,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 182–191

    2023

  38. [38]

    Decentralised variational inference frameworks for multi-object tracking on sensor networks,

    Q. Li, R. Gan, and S. J. Godsill, “Decentralised variational inference frameworks for multi-object tracking on sensor networks,”IEEE Trans- actions on Signal Processing, vol. 73, pp. 2753–2767, 2025. 1 Supplementary Materials for PiV oT: A Variational Solution for Real-time Large-scale Multi-object Detection and Tracking under Heavy Clutter APPENDIXA NOTA...

    2025

  39. [39]

    Forcingq(E k = 0) = 0means the objectkmust always exist, eliminating any meaningful uncertainty quantification of existence

  40. [40]

    Either outcome is problematic, making the naive mean-field assumption in (13) fundamentally flawed

    Otherwise, the objectkcannot associate with any mea- surement, preventing it from ever being updated. Either outcome is problematic, making the naive mean-field assumption in (13) fundamentally flawed. APPENDIXC OPTIMALNHPP APPROXIMATIONVIEW OF(22) This appendix provides a detailed interpretation of the approximation in (22) from the perspective of point ...

  41. [41]

    Point process interpretation of the approximation:Re- call from Section II-A that the overall measurement process is constructed by superposing a clutter process with inten- sityΛ 0ℓ0(Yj)andKindependent NHPPs, where thek-th NHPP has intensityD kΛkℓ(Yj|Xk, Pk), dependent on the detectability variableD k. From (4)-(7), the joint association and likelihood d...

  42. [42]

    Specifically, we consider the following result: 4 Lemma C.1.Letq(Z, m)denote the likelihood density of an NHPP

    Justification as KLD minimisation:We justify the em- ployed NHPP approximation by showing that it arises naturally as the solution to a KLD minimisation problem. Specifically, we consider the following result: 4 Lemma C.1.Letq(Z, m)denote the likelihood density of an NHPP . Among all NHPP likelihoods, theq(Z, m)that minimises the KLD from theD k-marginali...

  43. [43]

    Justification for well-separated objectk:Specifically, we show that the approximation error does not affect the Stage 2 inference for objectkwhereq 1(X,Λ, P)is fixed. The argument is to show that, under the same gating reduction that is safe for a sufficiently isolated object, the Stage 2 update without the approximation in (22) and the employed Stage 2 u...

  44. [44]

    Specifically, we may incorporateX k,Λ k, Pk inq 2 and exclude them from the fixedq 1

    Approximation mitigation and Stage 2 refinement:The approximation error from (22) introduced in Stage 1 for q1(X,Λ, P)can be mitigated or removed by extending the Stage 2 objective to include a refinement step. Specifically, we may incorporateX k,Λ k, Pk inq 2 and exclude them from the fixedq 1. In doing so, we continue to approximate the same target dist...

  45. [45]

    Computing allg(D k = 0) fork= 1, ..., Kn requires evaluating allg k j via either (67) or (68)

    Cost analysis:We now assess the cost in terms of expo- nential and logarithmic operations. Computing allg(D k = 0) fork= 1, ..., Kn requires evaluating allg k j via either (67) or (68). Specifically: 1) Each use of (67) involves one exponential and one logarithm, and is applied up toK nMtimes. 2) Each use of (68) involvesK n exponentials and one logarithm...

  46. [46]

    Building on this birth initialisation, we recommend first running two to three CA VI iterations for existing objectsk= 1,

    Variational initialisation and closely spaced birth prun- ing:Points (i)–(iii) of Section IV-D suggest using a flat factorised birth prior, initialising newly born components over distinct small positional regions, and tiling these regions over the surveillance area. Building on this birth initialisation, we recommend first running two to three CA VI iter...

  47. [47]

    The procedure is applied after inference, operates only on inferred track statistics, and does not modify the variational updates

    Post-processing details for Doppler PiVoT moving object selection:This appendix details the post-processing stage for moving-object selection used in the Doppler PiV oT automotive radar experiments. The procedure is applied after inference, operates only on inferred track statistics, and does not modify the variational updates. Its purpose is to extract a...

  48. [48]

    Lemma H.1.Let0< a≤α, and0< x≤1

    Preliminary algebraic lemmas:We begin by presenting algebraic lemmas that will be used in the subsequent analysis. Lemma H.1.Let0< a≤α, and0< x≤1. Then, 1− 1 1 +ax ≤ 1− 1 1 +a α+ 1 α+ 1/x .(69) Proof.Sincea+ 1≤a+ 1/x, addingα−a≥0to both the numerator and denominator of a+1 a+1/x does not decrease the ratio; hence, a+1 a+1/x ≤ α+1 α+1/x. Therefore, we have...

  49. [49]

    Lemma H.3.Suppose the prior covarianceΣ ′ k in(12), used in the state update(27), satisfiesHΣ ′ kH ⊤ =cI dY , wherec >0 is a scalar andI dY is the identity matrix

    Analysis of CAVI update:Next, we present two lemmas and a proposition concerning the employed CA VI updates and their bounding behaviour. Lemma H.3.Suppose the prior covarianceΣ ′ k in(12), used in the state update(27), satisfiesHΣ ′ kH ⊤ =cI dY , wherec >0 is a scalar andI dY is the identity matrix. Then the updated covarianceΣ k in(27)satisfiesHΣ kH ⊤ =...

  50. [50]

    Proof.Suppose PM j=1 q− 1 (θj =k)< s

    Proof of Theorem IX.1:With Proposition H.5, Corollary H.2.1, and Lemma H.1, we are now ready to prove Theo- rem IX.1. Proof.Suppose PM j=1 q− 1 (θj =k)< s. Proposition H.5 first gives the one-step upper bound (75) onq + 1 (θj =k). We further bound its right-hand side using Lemma H.1 withx= 1 dk(yj ), a=c(s) exp(− dY /2PM j=1 q− 1 (θj =k) ), andα=c(s) exp(...

  51. [51]

    c(s)Bk(s) dY /2 exp(− dY /2 Bk(s) )≥eand PM j=1 q− 1 (θj =k)< Vk(s), whereV k(s)is defined in (54), or

  52. [52]

    Recall that we have also assumed PM j=1 q− 1 (θj =k)< s

    c(s)Bk(s) dY /2 exp(− dY /2 Bk(s) )< e. Recall that we have also assumed PM j=1 q− 1 (θj =k)< s. Therefore, combining the above conditions gives the threshold Lk(s)in (53). Hence, if PM j=1 q− 1 (θj =k)< L k(s), thenPM j=1 q+ 1 (θj =k)< ∆k(s) Bk(s) PM j=1 q− 1 (θj =k). Since∆ k(s)≤B k(s)by the admissibility defined in (50), it follows that PM j=1 q+ 1 (θj...

  53. [53]

    The main message is twofold

    ThresholdL k(s)behaviour with respect tosandB k(s): Here we examine how the thresholdL k(s)in Theorem IX.1 changes withsand the admissible boundB k(s). The main message is twofold. First, for fixeds, a tighter/lower ad- missible boundB k(s)gives a no smaller thresholdL k(s). Second, the dependence onsis more subtle: decreasings 9 improves the internal bou...

  54. [54]

    In this case, we assume that the expected precision ofq 1(Pk) remains fixed at its prior value, namelyϕ kΦk =ϕ ′ kΦ′ k

    Construction ofB k(s)for the known-shape case:We consider the case where the object shape is known or the shape prior is sufficiently strong that the shape update is negligible. In this case, we assume that the expected precision ofq 1(Pk) remains fixed at its prior value, namelyϕ kΦk =ϕ ′ kΦ′ k. We now construct a tighter admissible boundB k(s)in (50) fo...

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.