Pith. sign in

REVIEW 3 major objections 7 minor 2 cited by

Joint Multitarget Detection and Tracking with mmWave Radar

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read An integrated detection-association-tracking pipeline with a Cramér-Rao-bound-based measurement model lets an mmWave radar track multiple weak targets in clutter.

desk verdict A coherent applied integration of NOMP, SPA, and KF with a real runtime win, but the weak-target claim rests on a CRB-efficiency assumption that only holds above 17 dB. read the letter →

arxiv 2412.17211 v2 pith:KD3ZWSDM submitted 2024-12-23 eess.SP

Jiang Zhu , Menghuai Xu , Ruohai Guo , Fangyong Wang , Guangying Zheng , Fengzhong Qu This is my paper · ORCID
classification eess.SP
keywords mmWaveradarmulti-targettrackingsuper-resolutionlinespectralestimationNewtonizedorthogonalmatchingpursuitsum-productalgorithmKalmanfilterCramér-Raoboundconstantfalsealarmrate
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a single integrated pipeline—super-resolution detection and estimation, sum-product data association, and a Kalman filter whose measurement noise covariance is set by the Cramér-Rao bound—can track multiple weak targets with an mmWave radar in cluttered scenes. The authors argue that their MNOMP-SPA-KF outperforms cascades built from FFT detection, PDA, JPDA, and conventional SPA, giving more stable trajectories, lower time-average MOSPA, and the ability to keep a weak target on track even when the detector alone misses it. They also report that the sensing step runs in about 0.075 seconds per frame, fast enough for the 0.1-second frame interval used in the experiments. The payoff, if correct, is that detectors can be run at higher false-alarm rates and rely on tracking to suppress clutter, improving weak-target detection without a full expensive three-dimensional spectral search.

What carries the argument

The load-bearing object is the pseudo-measurement model $z(t)=[p_x,p_y]^T+v(t)$ with $v(t)\sim \mathcal{N}(0,R(t))$ and $R(t)=\kappa\,\mathrm{CRB}([p_x,p_y]^T)$ evaluated at the 2D-MNOMP estimates, with $\kappa=1.2$. This turns line-spectral estimates into a standard measurement model and lets the Kalman filter and SPA gate use the elliptical uncertainty of the estimates instead of an isotropic approximation. A second mechanism is the 3D validation gate, which augments the two position coordinates with the estimated radial velocity using the augmented measurement matrix $H'_k$ and $R'_k(t)=\kappa\,\mathrm{CRB}([p_x,p_y,v]^T)$. A third mechanism is the forward-only 2D-MNOMP-CFAR with an invalid-target counter $K_{\mathrm{invalid}}$, which stops detection after a run of sub-threshold candidates and thereby removes the backward step's cost while limiting target masking.

What would settle it

Run 2D-MNOMP on a single target at integrated SNRs of 5, 10, 15 and 20 dB over hundreds of Monte Carlo trials and compare the empirical error covariance of the position estimates with $\kappa$ times the Cramér-Rao bound; if at 5 dB the empirical covariance is much larger or differently shaped than the scaled bound, then the pseudo-measurement covariance and the 3D gate are mis-specified at the weak-target operating point.

Watch

Extended reading notes

Core claim

The paper's central claim is that joint multi-target detection and tracking is best done by treating the mmWave radar's range-velocity-azimuth measurements as a pseudo-position measurement with a Cramér-Rao-bound-derived covariance, then feeding those measurements through radial-velocity-gated SPA and a Kalman filter. The proposed MNOMP-SPA-KF uses a forward-only 2D multisnapshot Newtonized orthogonal matching pursuit with a CFAR stopping rule and an invalid-target counter to avoid masking by strong targets; azimuth is obtained by least squares fit of the snapshot gains; the resulting position estimates are modeled as Gaussian with covariance $\kappa\,\mathrm{CRB}([p_x,p_y]^T)$. The SPA gate includes radial velocity as a third measurement dimension. The paper states that 3DNOMP-SPA-KF and MNOMP-SPA-KF generate the most stable trajectories and can accurately track the targets all the time in the simulated scenes, and that the pipeline detects a weak target that 2D-MNOMP alone misses.

Load-bearing premise

The load-bearing premise is that the detector's position estimates are accurate enough that their true error spread is well described by 1.2 times the theoretical best possible spread (the Cramér-Rao bound) evaluated at the estimate; the paper demonstrates this only at integrated SNR above about 17 dB, while its weak-target demonstration runs at 5.2 dB.

Editorial extensions

If this is right

  • Higher detection-stage false-alarm rates become usable, because the SPA-KF tracking stage suppresses most false alarms while preserving the weak target.
  • Targets separated by less than the nominal range resolution (0.5 m apart, versus 0.78 m resolution) can still be resolved and tracked as distinct objects thanks to the super-resolution MNOMP detector.
  • Using the scaled Cramér-Rao covariance in the Kalman filter outperforms using a scaled identity covariance, and adding radial velocity to the SPA gate outperforms the position-only 2D gate, in both simulations and real experiments.
  • The forward-only 2D-MNOMP drops the per-frame sensing time from about 1.69 seconds to about 0.075 seconds, making the pipeline compatible with the 0.1-second frame interval in the experiments.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The transferable idea is probably the elliptical pseudo-measurement covariance itself: any super-resolution estimator with a computable Cramér-Rao bound could be plugged into the same SPA-KF architecture, and the paper's comparisons suggest the gain comes more from using the true error shape than from the specific detector.
  • The weak-target gain in the paper's Section IV-D likely comes mostly from temporal integration by the tracker (gating, extrapolation, and track maintenance) rather than from improved single-frame detection; a per-frame detection-probability comparison before and after tracking would make that division explicit.
  • The radial-velocity gate should be tested at the same low-SNR operating point used for the weak target, because the velocity estimate is part of the gate and its error also grows as SNR drops; the 3D gate's advantage over the 2D gate may shrink or reverse there.
  • The constant-velocity motion model and fixed extrapolation count are simple track-management choices; the rest of the pipeline would survive replacing them with a maneuver-aware model, which is an extension the paper does not explore.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. This paper proposes an end-to-end detection–association–tracking pipeline for mmWave LFMCW radar. The detection module is a low-complexity forward-only 2D multisnapshot Newtonized orthogonal matching pursuit with CFAR (2D-MNOMP), which extracts range, Doppler, and azimuth; the association module is a sum-product algorithm whose validation gate includes radial velocity; the tracking module is a Kalman filter whose measurement covariance is set to a scaled Cramér–Rao bound (Proposition 1, Eq. (17)). The paper validates the CRB approximation in simulation, compares the integrated MNOMP-SPA-KF with FFT/PDA/JPDA/SPA variants on MOSPA, demonstrates a weak-target tracking scenario, and reports three real-data experiments with an AWR1642 radar. The central claim is that the integrated framework, especially with the 3D radial-velocity gate, produces the most stable trajectories and can track targets that 2D-MNOMP alone misses.

Significance. If the claims hold, the paper offers a useful engineering framework: the CRB-based pseudo-measurement covariance is a principled way to capture range–angle measurement correlation, the forward-only NOMP with a masking countermeasure and a single FFT is a meaningful complexity reduction, and the radial-velocity 3D gate is a sensible extension of SPA data association. The reported per-frame runtime of 0.075 s is attractive for real-time mmWave radar tracking. The paper also gives a closed-form CRB derivation with a proof, which is a concrete theoretical contribution. However, the evaluation currently leaves open whether the CRB-efficiency assumption holds in the weak-target regime, and the numerical and real-data comparisons lack the statistical or ground-truth support needed to substantiate the strongest claims.

major comments (3)
  1. [IV-D and III-A (Eqs. (21), (34c))] The advertised weak-target advantage is not quantitatively supported. Fig. 5 shows that 2D-MNOMP estimates approach the CRB only above roughly 17 dB integrated SNR, yet the weak-target scenario in Section IV-D uses an integrated SNR of 5.2 dB, about 12 dB below that threshold. In this regime the pseudo-measurement model (21) with R(t)=kappa CRB, the 3D gate (34c), and the Kalman gain (38) are likely misspecified because the estimator is probably biased and non-Gaussian. Moreover, Section IV-D reports no metric (e.g., track-loss probability, MOSPA, detection probability), and Fig. 8 shows a single realization; the maintained trajectory may be due to extrapolation (Section III-D) rather than to valid detections. Please add quantitative weak-target experiments, including the empirical covariance of 2D-MNOMP versus kappa CRB at 5.2 dB, and report detection probability and track-loss metrics averaged over many trials.
  2. [IV-C, Fig. 7] The central MOSPA comparison is averaged over 300 Monte Carlo trials but is plotted without error bars, confidence intervals, or statistical tests. The trajectory plots in Fig. 6 are single realizations, so the reader cannot assess whether the reported advantage of MNOMP-SPA-KF and 3DNOMP-SPA-KF over, e.g., MNOMP-SPA-subKF is significant or seed-dependent. Please report means with standard errors or boxplots, state whether all algorithms process identical scene realizations, and give the number of independent trials underlying each curve.
  3. [V, Figs. 11-17] The real-data evaluation is purely qualitative. The text asserts that MNOMP-SPA-KF (3D-Gate) tracks people while suppressing false alarms, but no ground-truth comparison (e.g., hand-labeled positions, GPS, or optical reference), no RMSE, and no track completeness or false-track statistics are presented. Since Section V is used to support the claim of superior real-world performance, please add quantitative evaluation: trajectory error against ground truth, track detection rate, false track rate, and a direct 2D-Gate versus 3D-Gate comparison using these metrics.
minor comments (7)
  1. [Abstract] The abstract contains typographical and grammatical errors, including 'Newtonalized' for 'Newtonized' and the phrase 'create smart, efficient, automated system'; these should be corrected.
  2. [III-A, Eq. (14)] The notation F_omega(Y_l) is used before its definition; please define the 2D FFT convention explicitly and state over which frequency grid it is evaluated.
  3. [III-A, Eq. (15), Abstract] The abstract describes the azimuth extraction as a conventional Bartlett beamformer, but Eq. (15) is a least-squares fit of the gain vector under the constraint (9); the terminology should be aligned with the actual computation.
  4. [Table I] The real-data experiments use L=4 receivers while the simulations use L=8; please comment on how this difference affects the CRB and the expected performance of the 3D gate.
  5. [III-D, Eq. (43)] The new-target initialization in Eq. (43) implicitly sets the lateral velocity to zero by aligning velocity with the radial direction; this assumption should be stated explicitly, since it may bias tracking of crossing targets.
  6. [IV-A, V] Several hyperparameters (K_invalid, N_ext, P_FA, PG, kappa, SPA iteration count) are set to fixed values without sensitivity analysis or a default-selection procedure; a short sensitivity study or a stated selection rule would improve reproducibility.
  7. [V, runtime paragraph] The runtime comparison reports 1.69 s versus 0.075 s per frame for the detector alone, and separately reports total runtimes for the full pipeline; please clarify the hardware/software environment and whether the 0.075 s figure includes SPA and KF or only the detector.

Circularity Check

0 steps flagged · score 0.0 of 10

There is no significant circularity: the CRB model is derived from the measurement equation and the tracking metric is external to the fitted κ parameter.

full rationale

The paper's derivation chain is self-contained. The CRB in Proposition 1 and eq. (67) is obtained from the baseband measurement model (1) through standard Fisher-information and reparameterization arguments (Lemmas 1-2), not from the tracking output. The pseudo-measurement covariance R(t)=κCRB(...) in eq. (21) is an approximation assumption about the 2D-MNOMP estimator; Fig. 5 directly checks estimator MSE against the CRB, and κ=1.2 is calibrated to the empirical coverage in Fig. 3. This is parameter calibration, not a prediction that reduces to the fit, because the performance metric (MOSPA, eq. (46)) is computed against ground truth and is independent of how κ was chosen. The SPA validation gate (33) and 3D-Gate (34c) use the same CRB model as part of the algorithm design, not as a post-hoc justification. The only self-citation, ref. [11], supplies the underlying NOMP-CFAR algorithm; it is a published prior result and does not by itself force the integrated tracking conclusions. The weak-target demonstration (Section IV-D) relies on extrapolation across missed detections; that is temporal filtering behavior, not a tautology. A genuine robustness concern exists—CRB efficiency is demonstrated only for SNR ≳17 dB while the weak-target example operates at 5.2 dB—but this is a validity limitation, not a circularity.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a standard Gaussian line-spectrum measurement model, a constant-velocity motion model, and a strong statistical assumption that the suboptimal 2D-MNOMP estimator is CRB-efficient in the operating regime. The kappa scaling of the CRB is a fitted parameter, and several hyperparameters are chosen by hand. No new physical entities are introduced.

free parameters (8)
  • kappa (CRB scaling) = 1.2
    Scales the Cramér-Rao bound to form the pseudo-measurement covariance R(t)=kappa CRB; calibrated to the simulation in Fig. 3 so 95 percent of 2D-MNOMP estimates fall in the ellipse, then reused in all tracking runs.
  • K_invalid = not stated
    Stops the 2D-MNOMP forward search after K_invalid consecutive sub-threshold detections; example value 2 given in text, but the value used in experiments is not stated.
  • N_ext = not stated
    Number of consecutive missed detections before a target is deleted; introduced in Section III-D but no numerical value is given.
  • P_FA = 0.01/(128x64) or 0.1/(128x64)
    False alarm probability of the CFAR detector; set to 0.01/(128x64) in most experiments and 0.1/(128x64) in the weak-target simulation.
  • CFAR training and guard cells = [5,4] training, [3,3] guard; 60 training cells in real data
    Number of training and guard cells for the CFAR detector; [5,4] training bands and [3,3] guard bands in simulation, 60 training cells in real data.
  • Gate probability PG = 0.95
    Probability used to compute the validation gate threshold dG; set to 0.95.
  • SPA iterations = 10
    Number of message-passing iterations in the sum-product algorithm; set to 10.
  • K_max = 30
    Upper bound on the number of targets for the detector; set to 30.
assumptions (5)
  • domain assumption Received signal follows the point-target Gaussian model in eq. (1), with i.i.d. complex Gaussian noise across fast-time, slow-time, and spatial domains.
    The entire MNOMP estimator and CRB derivation rest on this model; real radars include multipath, mutual coupling, and extended targets that are not modeled.
  • domain assumption Targets follow a constant-velocity motion model with known process noise covariance Q (eq. 3).
    The Kalman filter and SPA prediction rely on this model; pedestrians and vehicles do not always move at constant velocity.
  • ad hoc to paper The 2D-MNOMP estimator is asymptotically unbiased and CRB-efficient, so its error covariance equals the single-target CRB (eq. 18-21).
    Assumed without proof for a greedy suboptimal algorithm; Fig. 5 validates CRB proximity only for SNR above about 17 dB, while the weak-target experiment uses 5.2 dB integrated SNR.
  • domain assumption Data association is one-to-one between targets and measurements, with i.i.d. clutter modeled by the distribution fc (eq. 23-27).
    The SPA factor graph assumes one measurement per target and one target per measurement; extended targets generate multiple measurements, so the paper clusters them for real data, which is an unmodeled preprocessing step.
  • ad hoc to paper New targets are initialized with velocity aligned with the radial direction (eq. 43).
    The new track state is set using only the radial velocity components, assuming purely radial motion; this biases tracks for targets moving tangentially.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Joint Multitarget Detection and Tracking with mmWave Radar." pith.science (2026). https://pith.science/paper/KD3ZWSDM

@misc{pith2026241217211,
  author       = {Pith},
  title        = {Pith review of: Joint Multitarget Detection and Tracking with mmWave Radar},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KD3ZWSDM}},
  note         = {Machine review of arXiv:2412.17211}
}
read the original abstract

Accurate targets detection and tracking with mmWave radar is a key sensing capability that will enable more intelligent systems, create smart, efficient, automated system. This paper proposes an end-to-end detection-estimation-track framework named MNOMP-SPA-KF consisting of the target detection and estimation module, the data association (DA) module and the target tracking module. In the target estimation and detection module, a low complexity, super-resolution and constant false alarm rate (CFAR) based two dimensional multisnapshot Newtonalized orthogonal matching pursuit (2D-MNOMP) is designed to extract the multitarget's radial distances and velocities, followed by the conventional (Bartlett) beamformer to extract the multitarget's azimuths. In the DA module, a sum product algorithm (SPA) is adopted to obtain the association probabilities of the existed targets and measurements by incorporating the radial velocity information. The Kalman filter (KF) is implemented to perform target tracking in the target tracking module by exploiting the asymptotic distribution of the estimators. To improve the detection probability of the weak targets, extrapolation is also coupled into the MNOMP-SPA-KF. Numerical and real data experiments demonstrate the effectiveness of the MNOMP-SPA-KF algorithm, compared to other benchmark algorithms.

Figures

Figures reproduced from arXiv: 2412.17211 by the authors.

Figure 1
Figure 1. The typical measurement setup of the LFMCW mmWave radar. For simplicity, both the target index and the time index [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. The flow chart of MNOMP-SPA-KF algorithm. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. The estimates of weak target in multitargets scene output by 2D-MNOMP and the confidence ellipse evaluated by the [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Factor graph for multi-targets data association in a single mmWave radar. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: The CRB in eq.(17) and the estimated MSE by 3D-NOMP and 2D-MNOMP. (a) MSE and CRB of [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: The tracking results of different algorithms. The arrow indicates the direction of the target movement. [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: The time-average MOSPA of different algorithms. [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: The results of MNOMP and MNOMP-SPA-KF in the scene where two strong targets and one weak target coexist. (a) [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]
Figure 9
Figure 9. Figure 9: Physical image of AWR1642 radar. TABLE I PARAMETERS SETTING OF THE EXPERIMENT Parameters Value Carrier frequency fc 77 GHz Frequency modulation slope µ 8.012 MHz/µs sweep time Tp 56µs Pulse repeat interval Tr 3µs Bandwidth B 448.672 MHz Sampling frequency fs = 1/Ts 5 M…
Figure 10
Figure 10. Figure 10: The scene and original state of experiment I. [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]
Figure 11
Figure 11. Figure 11: The tracking results of different algorithms. The arrow indicates the direction of the target movement. [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]
Figure 12
Figure 12. Figure 12: The scene and original state of experiment II. [PITH_FULL_IMAGE:figures/full_fig_p027_12.png]
Figure 13
Figure 13. Figure 13: The tracking results of experiment II by different algorithms. The arrow indicates the direction of the target movement. [PITH_FULL_IMAGE:figures/full_fig_p027_13.png]
Figure 14
Figure 14. Figure 14: The tracking results of experiment II using 2D-Gate and 3D-Gate in [PITH_FULL_IMAGE:figures/full_fig_p028_14.png]
Figure 15
Figure 15. Figure 15: The scene and original state of experiment III. [PITH_FULL_IMAGE:figures/full_fig_p029_15.png]
Figure 16
Figure 16. Figure 16: The tracking results of experiment III by different algorithms. The arrow indicates the direction of the target movement. [PITH_FULL_IMAGE:figures/full_fig_p029_16.png]
Figure 17
Figure 17. Figure 17: The tracking result of experiment III using 2D-Gate and 3D-Gate in [PITH_FULL_IMAGE:figures/full_fig_p030_17.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. mmFlux: Crowd Flow Analytics with Commodity mmWave MIMO Radar

    eess.SP 2025-07 conditional novelty 6.0 of 10

    mmFlux estimates a crowd flow field from commodity mmWave radar point clouds, reconstructs a directed graph of splits and merges, and uses divergence and curl to identify gatherings, dispersions, and turns.

  2. Message Passing for Track-Before-Detect

    eess.SP 2025-06 conditional novelty 5.0 of 10

    A scalable belief-propagation track-before-detect method tracks multiple objects directly from correlated raw sensor data with fluctuating amplitudes and unknown noise, outperforming detect-then-track baselines.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages · cited by 2 Pith papers

  1. [1]

    Design Guide: TIDEP-01000 People Tracking and Counting Reference Design Using mmWave Radar Sensor

    work page

  2. [2]

    Multiple-target tracking and track management for an FMCW radar network,

    D. B. Kim, S. M. Hong, “Multiple-target tracking and track management for an FMCW radar network,” EURASIP J. Adv. Signal Process. , 2013

    work page 2013

  3. [3]

    People Tracking and Counting Reference Design Using mmWave Radar Sensor

    Texas Instrument, “People Tracking and Counting Reference Design Using mmWave Radar Sensor” [Online]. Available: https://www.ti.com/lit/ug/tidue71d/tidue71d.pdf

    work page

  4. [4]

    Create an Intelligent Building System: Use mmWave to Count and Track People

    Texas Instrument, “Create an Intelligent Building System: Use mmWave to Count and Track People” [Online]. Available: https://www.ti.com.cn/cn/lit/ta/sszt725/sszt725.pdf

    work page

  5. [5]

    DFT-based estimation of damped oscillation parameters in low-frequency mechanical spectroscopy,

    K. Duda, L. B. Magalas, M. Majewski, and T. P. Zielinski, “DFT-based estimation of damped oscillation parameters in low-frequency mechanical spectroscopy,” IEEE Trans. Instrum. Meas. , vol. 60, no. 11, pp. 3608-3618, 2011

    work page 2011

  6. [6]

    M. A. Richards, Fundamentals of Radar Signal Processing , Second edition. New York: McGraw- Hill Education, 2014

    work page 2014

  7. [7]

    Sensitivity of basis mismatch to compressed sensing,

    Y . Chi, L. L. Scharf, A. Pezeshki and R. Calderbank, “Sensitivity of basis mismatch to compressed sensing,” IEEE Trans. Signal Process. , vol. 59, pp. 2182-2195, 2011

    work page 2011

  8. [8]

    Variational Bayesian inference of line spectra,

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

    work page 2017

Show all 34 references
  1. [9]

    Atomic norm denoising with applications to line spectral estimation,

    B. N. Bhaskar, G. Tang, and B. Recht, “Atomic norm denoising with applications to line spectral estimation,” IEEE Trans. Signal Process. , vol. 61, no. 23, pp. 5987-5999, 2013. September 25, 2025 DRAFT 35

    2013

  2. [10]

    Newtonized orthogonal matching pursuit: frequency estimation over the continuum,

    B. Mamandipoor, D. Ramasamy, and U. Madhow, “Newtonized orthogonal matching pursuit: frequency estimation over the continuum,” IEEE Trans. Signal Process. , vol. 64, no. 19, pp. 5066- 5081, 2016

    2016

  3. [11]

    CFAR based NOMP for line spectral estimation and detection

    M. Xu, J. Zhu, J. Fang, N. Zhang and Z. Xu, “CFAR based NOMP for line spectral estimation and detection”, IEEE Trans. Aerosp. and Electron. Syst. , vol. 59, no. 5, pp. 6971-6990, 2023

    2023

  4. [12]

    Challa, M

    S. Challa, M. R. Morelande, D. Mu ˇsicki, and R. J. Evans, Fundamentals of Object Tracking, 2011, Cambridge, U.K.: Cambridge University Press

    2011

  5. [13]

    An optimal tracking filter for processing sensor data of imprecisely determined origin in surveillance systems,

    R. A. Singer and J. J. Stein, “An optimal tracking filter for processing sensor data of imprecisely determined origin in surveillance systems,” 1971 IEEE Conference on Decision and Control , pp. 171-175, 1971

    1971

  6. [14]

    Tracking in a cluttered environment with probabilistic data association,

    Y . Bar-Shalom and E. Tse, “Tracking in a cluttered environment with probabilistic data association,” Automatica, vol. 11, no. 3, pp. 451-460, 1975

    1975

  7. [15]

    Sonar tracking of multiple targets using joint probabilistic data association,

    T. Fortmann, Y . Bar-Shalom and M. Scheffe, “Sonar tracking of multiple targets using joint probabilistic data association,” IEEE J. Ocean. , vol. 8, no. 3, pp. 173-184, 1983

    1983

  8. [16]

    trackerJPDA

    MATLAB function “trackerJPDA”, available at https://jp.mathworks.com/help/fusion/ref/trackerjpda- system-object.html, 2024

    2024

  9. [17]

    Density-based spatial clustering of applications with noise,

    M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, “Density-based spatial clustering of applications with noise,” in Proc. Int. Conf. Knowl. Discovery Data Mining , vol. 240, 1996, p.6

    1996

  10. [18]

    Approximate evaluation of marginal association probabilities with belief propagation,

    J. L. Williams and R. A. Lau, “Approximate evaluation of marginal association probabilities with belief propagation,” IEEE Trans. Aeros. and Electron. Syst. , vol. 50, no. 4, pp. 2942-2959, 2014

    2014

  11. [19]

    A scalable algorithm for tracking an unknown number of targets using multiple sensors,

    F. Meyer, P. Braca, P. Willett, and F. Hlawatsch, “A scalable algorithm for tracking an unknown number of targets using multiple sensors,” IEEE Trans. Signal Process. , vol. 65, no. 13, pp. 3478- 3493, 2017

    2017

  12. [20]

    Message passing algorithms for scalable multitarget tracking,

    F. Meyer et al., “Message passing algorithms for scalable multitarget tracking,” Proc. IEEE, vol. 106, no. 2, pp. 221-259, 2018

    2018

  13. [21]

    Scalable detection and tracking of geometric extended objects

    F. Meyer and J. L. Williams, “Scalable detection and tracking of geometric extended objects”, IEEE Trans. Signal Process., vol. 69, pp.6283-6298, 2021

    2021

  14. [22]

    Self-tuning algorithms for multisensor-multitarget tracking using belief propagation

    G. Soldi, F. Meyer, P. Braca, and F. Hlawatsch, “Self-tuning algorithms for multisensor-multitarget tracking using belief propagation”, IEEE Trans. Signal Process. , vol. 67, no. 15, pp. 3922-3937, 2019

    2019

  15. [23]

    Neural enhanced belief propagation for multiobject tracking

    M. Liang and F. Meyer, “Neural enhanced belief propagation for multiobject tracking”, IEEE Trans. Signal Process., vol. 72, pp. 15-30, 2024

    2024

  16. [24]

    Introduction to mmwave Sensing: FMCW Radars

    Texas Instrument, “Introduction to mmwave Sensing: FMCW Radars.” [Online]. Available: September 25, 2025 DRAFT 36 https://training.ti.com/sites/default/files/docs/mmwaveSensing-FMCW-offlineviewing 0.pdf

    2025

  17. [25]

    Bar-Shalom, P

    Y . Bar-Shalom, P. K. Willet and X. Tian, Tracking and Data Fusion , 2011, YBS Publishing

    2011

  18. [26]

    S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation Theory , 1993, Englewood Cliffs, NJ: Prentice Hall

    1993

  19. [27]

    Max sum based data associations for tracking point and extended Targets,

    W. Ma, Z. Jing, P. Dong and H. Leung, “Max sum based data associations for tracking point and extended Targets,” IEEE Trans. Aeros. and Electron. Syst. , doi: 10.1109/TAES.2024.3482287

    2024

  20. [28]

    An introduction to factor graphs,

    H.-A. Loeliger, “An introduction to factor graphs,” IEEE Signal Process. Mag. , vol. 21, no. 1, pp. 28-41, 2004

    2004

  21. [29]

    Bar-Shalom, X

    Y . Bar-Shalom, X. R. Li, and T. Kirubarajan, Estimation with applications to tracking and navigation: theory, algorithms and software , 2004, John Wiley & Sons

    2004

  22. [30]

    The probabilistic data association filter,

    Y . Bar-Shalom, F. Daum and J. Huang, “The probabilistic data association filter,” IEEE Control Systems Magazine, vol. 29, no. 6, pp. 82-100, 2009

    2009

  23. [31]

    A metric for performance evaluation of multi-Target tracking algorithms,

    B. Ristic, B. -N. V o, D. Clark and B. -T. V o, “A metric for performance evaluation of multi-Target tracking algorithms,” IEEE Trans. Signal Process. , vol. 59, no. 7, pp. 3452-3457, 2011

    2011

  24. [32]

    S. M. Kay, Fundamentals of Statistical Signal Processing: Detection Theory , 1993, Englewood Cliffs, NJ: Prentice Hall

    1993

  25. [33]

    Iovescu and S

    C. Iovescu and S. Rao, The fundamentals of millimeter wave sensors, Texas Instruments, pages 1–8, 2017

    2017

  26. [34]

    Huang, Indoor Localization and Mapping with 4D mmWave Imaging Radar, Master Thesis, available at https://www.ri.cmu.edu/app/uploads/2024/05/MSR thesis juite-1.pdf

    J. Huang, Indoor Localization and Mapping with 4D mmWave Imaging Radar, Master Thesis, available at https://www.ri.cmu.edu/app/uploads/2024/05/MSR thesis juite-1.pdf. September 25, 2025 DRAFT

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.