Pith. sign in

REVIEW 6 major objections 6 minor 35 references

Distant Object Localisation from Noisy Image Segmentation Sequences

T0 review · 6 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read Distant objects localised in 3D from noisy image segments and camera poses alone, using a bootstrap particle filter

desk verdict A standard particle filter applied to a niche problem works for static targets, but the wildfire-smoke framing is unsupported by the motion model. read the letter →

arxiv 2509.20906 v3 pith:HJPBHVHI submitted 2025-09-25 cs.CV cs.RO

Julius Pesonen , Arno Solin , Eija Honkavaara This is my paper
classification cs.CVcs.RO
keywords particlefilter3Dlocalisationimagesegmentationdronesurveillancewildfiremonitoringnoisycameraposesmulti-targettracking
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

The paper claims that a distant static object can be localised in three dimensions from a moving camera using only a sequence of noisy image-segmentation masks and noisy camera poses, without any 3D reconstruction or feature matching. It shows that a bootstrap particle filter, which projects particles into the image and weights them by distance to segmented pixels, can converge to the object's position and also estimate shape and uncertainty. Simulation experiments place errors around 40–80 metres at roughly two kilometres, and a real drone sequence localises a mast at about 700 metres with a mean error near 92 metres. If right, this gives a lightweight, detector-agnostic localisation component for drone-based wildfire monitoring and similar surveillance tasks where depth sensors and reconstruction are impractical.

What carries the argument

The bootstrap particle filter (a Monte Carlo filter where the target's 3D position is a set of weighted particles) is the central mechanism. Each particle is projected through the pinhole camera model into the image plane, and its weight is updated by an exponential of the squared pixel distance to the nearest positive segmented pixel; resampling occurs at every observation, and the prediction step injects distance-scaled Gaussian noise. The filter is initialised from a least-squares estimate of the midpoint between two camera rays. For multiple targets, additional filters are spawned when segments appear that are far from the existing particle projections, with a dynamic threshold, and filt

What would settle it

A decisive experiment: reproduce the paper's simulation for a static 100-metre cube at 2 km with the reported pose noise, false-positive, and false-negative settings, and check whether the converged RMSE falls in the 40–80 m range. Separately, run the same filter on a target moving at a constant 0.5 m/s across the field of view; if the filter's RMSE does not track the moving ground truth, the static-assumption limit is confirmed.

Watch

Extended reading notes

Core claim

The central claim is that 3D localisation of distant objects can be solved as a Bayesian filtering problem using particle filters, rather than as a geometry or reconstruction problem. The filter represents the target's unknown 3D position as a cloud of particles; each observation projects the particles into the camera frame and re-weights them according to their pixel distance from the segmented positive pixels, with Gaussian noise added in the prediction step to account for uncertainty. The proposal is shown to handle multiple targets by running several filters spawned when out-of-distribution segments appear, and to tolerate realistic pose noise, false-positive segments, and false-negative

Load-bearing premise

The load-bearing premise is that the target object is static: the filter's prediction step only adds Gaussian diffusion and has no motion model, so a moving target (such as drifting wildfire smoke) could break convergence.

Editorial extensions

If this is right

  • If the central claim holds, a drone can localise a static distant object (e.g., a fire's smoke column base) in geographic coordinates using only its own GNSS pose estimates and the output of an existing segmentation network, with no cloud processing or 3D scene reconstruction.
  • The method is detector-agnostic: any segmentation model producing binary masks can be plugged in, so the same localisation code transfers across tasks.
  • The filter simultaneously estimates 3D position and an uncertainty cloud, which gives a principled way to report confidence or trigger revisits.
  • Multiple simultaneous targets can be tracked by independent filters, with targets appearing and disappearing over time.
  • Because the method runs on drone-carried compute with modest overhead (100k–1M particles), it is feasible for real-time on-edge deployment.

Reading between the lines

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

  • If the static-target assumption were relaxed to include slow drift, the same filtering machinery could be extended with a motion model in the prediction step (e.g., random-walk or constant-velocity velocity states); the paper's own convergence results suggest this is the natural next step, though the authors do not propose it.
  • The RMSE results degrade notably with false-positive segments near the target and with partial false negatives, which suggests the practical bottleneck may be segmentation quality near the object silhouette rather than pose accuracy; a testable extension is to compare the filter's performance against a mask-quality metric and to fuse multiple segmentation models.
  • The 92-metre real-sequence error at ~700 metres (about 13% relative) is achieved with uncalibrated camera intrinsics and ignored lever-arm/boresight offsets, implying that a small amount of physical calibration could substantially tighten the estimate—a cheap experiment that follows directly from their analysis.
  • The abstract mentions multi-view triangulation as an alternative, but the full text concentrates on the particle filter; a direct head-to-head on the same noisy inputs would clarify when filtering actually beats the geometric baseline.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper addresses 3D localisation of distant, static objects from a sequence of binary image segmentation masks and noisy camera poses, motivated by drone-based wildfire monitoring. The authors propose a bootstrap particle filter that projects particles into the camera frame, weights them against positive segmentation pixels, and resamples at every step. A multi-filter extension is introduced for multiple targets. The method is evaluated in a simulated pinhole-camera environment with injected pose noise, false positives, and false negatives, and on one real drone sequence of a telecommunication mast. The paper reports RMSE/NLPD metrics and concludes that the proposed filter, combined with a pre-existing segmentation model, can solve wildfire geolocalisation tasks.

Significance. If the empirical claims hold, the paper would provide a lightweight, detector-agnostic component for localising distant objects without feature correspondence or 3D reconstruction, which is relevant for edge-device drone applications. Strengths include the open-code release, a flexible simulation framework covering several realistic noise sources, and a real-data demonstration. However, the current evidence is not yet sufficient to establish the contribution: there are no baseline comparisons, hyperparameters are tuned on the evaluation scenarios without held-out validation, the filter has no motion model while the motivating application (wildfire smoke) is moving, and the reported simulation metrics lack uncertainty estimates. These gaps are load-bearing for the central claims.

major comments (6)
  1. [Abstract, §I, §III–IV] The abstract and introduction state that the task can be solved with 'either multi-view triangulation or particle filters', but no triangulation method is described or evaluated anywhere in §III–IV. Since triangulation is the natural baseline and is presented as part of the proposed solution, the paper should either report triangulation results on the same data or remove it from the claims. As written, the scope mismatch undermines the stated contribution.
  2. [§IV-A, Table I] No baselines are reported. Table I compares the particle filter only against itself under accumulating noise. Without comparison to simple line-of-sight intersection, least-squares triangulation, or a midpoint estimate, the results do not demonstrate that the particle filter is a necessary or better solution; a much simpler estimator might produce comparable RMSE. Please add baseline comparisons on the same simulation scenarios and the same real sequence.
  3. [§IV-B] Parameters (τ_min_obs, SD_init, T_s, N_p, θ_p−o, n_θ_dm, n_θ_fuse) were chosen 'through trial and error' on the same simulated configurations used for evaluation, with no held-out validation. Because the reported RMSE/NLPD values are produced on those very scenarios, the empirical demonstration is at risk of overfitting. Provide a separate validation set or a sensitivity analysis, and show that the conclusions are not an artifact of parameter tuning.
  4. [§II, §III-B, §V] The prediction step in §III-B injects only Gaussian diffusion; the filter has no velocity or motion model, and §II explicitly assumes a static target. Yet the abstract and conclusion extend the claim to drone-based wildfire monitoring, while wildfire smoke drifts with wind. All simulation and real experiments use static targets. This is a correctness-risk concern: a diffusion-only filter can systematically lag a moving plume or fail to converge. Either add a moving-target scenario (e.g., constant-velocity smoke or a drifting target) with appropriate motion modeling, or substantially temper the claims to static objects such as towers.
  5. [Table I, §IV-A, §IV-C] All simulation metrics are means over only ten runs, with no standard deviations, confidence intervals, or quantiles. With n=10, the differences between noise configurations may not be statistically meaningful. The real-data result in §IV-C similarly reports a mean over ten random seeds without showing the spread. Please report error bars or the full distribution of runs.
  6. [Eq. (10), §III-D] The NLPD definition is incomplete. Equation (10) writes 'p(m_t | p)' but does not define this density; if it is a normal density evaluated at the target mean using the particle distribution, the covariance estimate and bandwidth are unspecified, and the expression as printed is not a proper predictive log-density. This makes the uncertainty-quality claim non-reproducible. Define the exact computation.
minor comments (6)
  1. [Table I, rows 3–4] The RMSE (200–1000 m) and NLPD values for rows 3 and 4 are identical to two decimals despite different noise settings. Check whether this is a typo or an indication that the runs are not independent.
  2. [Figure 3 caption] The caption says 'RMSEs of the single target simulation experiments', but the figure shows multitarget results. The wording should be corrected to 'multitarget'.
  3. [§III-C] The parameter n_θ_fuse is listed but no fusion procedure is described in the algorithm or experiments. Clarify how and when filters are fused, or remove the parameter from the default set.
  4. [§IV-C] The empirical section states a mean RMSE of 92.00 m after 50 m of translation and a minimum mean RMSE of 76.88 m, but does not give standard deviation or range over the ten random seeds. A point estimate alone is insufficient.
  5. [Abstract/full-text mismatch] The abstract in the paper header mentions 'multi-view triangulation or particle filters', while the full-text abstract says 'using particle filters for both single and multiple target scenarios'. These should be harmonized.
  6. [Eq. (8)] The weight update uses the notation 'min((obs − p_proj)^2)'. Please define the norm and specify whether the minimum is taken over all positive pixels in the observation set; the formula as written is ambiguous.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the particle filter localisation is an empirical estimation procedure, not a prediction constructed from its own outputs.

full rationale

The paper's derivation chain is an estimation pipeline, not a self-referential construction. The particle filter initialises from a two-view least-squares midpoint (Section III-B-a), predicts by adding Gaussian diffusion (Section III-B-b), updates weights by comparing projected particles to positive segmentation pixels (Section III-B-c), and resamples (Section III-B-d). Each step consumes the actual inputs—camera poses and segmentation masks—and produces a posterior over target position. No equation defines the target location in terms of the filter output, and no fitted parameter is renamed as a prediction. The simulation is a controlled validation where the target position generates the observations, but the filter does not receive the ground-truth position; the real-data experiment likewise uses the known mast geolocation only for evaluation, not as a filter input. The parameter choices in Section IV-B ('Through trial and error with various experiment configurations, we decided on a reliable set of default parameters') and the adjustment of T_s and N_p on the real sequence are a potential overfitting/validity concern, not circularity: the reported RMSEs are measured outcomes, not algebraically forced by the parameter settings. The self-citation [1] merely motivates the wildfire application and is not load-bearing for the localisation derivation. The static-target assumption is an explicit modelling limitation, not a circular step. Therefore no significant circularity is present.

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

The paper introduces no new physical entities, forces, or dimensions; the multiple-filter architecture is a known technique. The ledger instead lists eight tuned hyperparameters and eight modeling assumptions that the central claim rests on.

free parameters (8)
  • Initialization distribution scale SD_init / P_init = SD_init = 1000; P_init covariance form not fully reported
    Set by trial and error so the initial particle cloud contains the target while remaining dense enough to converge; Section IV-B.
  • Translation step size T_s = 10 m (simulation), 2 m (empirical)
    Controls how often filter updates are triggered; chosen manually; too small harms convergence (Section IV-B, IV-C).
  • Number of particles N_p = 100000 (simulation), 1000000 (empirical)
    100k was unstable for the thin mast target, so it was raised to 1M after observing instability (Section IV-C).
  • Minimum consecutive observations to initialize a filter tau_min_obs = 5
    Trades false-positive initialization against speed of response; set by trial and error (Section IV-B).
  • Out-of-distribution threshold theta_p_o = 1 standard deviation
    Determines when remaining pixels spawn a new filter; chosen by trial and error (Section III-C, IV-B).
  • Filter dismissal threshold n_theta_dm = 5
    Number of observation steps without positives before a filter is removed; set by trial and error (Section III-C, IV-B).
  • Filter fusion threshold n_theta_fuse = 5
    Used for merging filters, details not given; set by trial and error (Section IV-B).
  • Prediction-step noise constant = not reported
    Particles are injected with Gaussian noise scaled by distance and an optimised constant; the constant value is not stated (Section III-B).
assumptions (8)
  • domain assumption Targets are static.
    Explicitly assumed in Section II; the prediction step has no velocity model, so moving targets such as drifting smoke are not handled.
  • domain assumption Camera pose sequence is available from GNSS/IMU with errors in the stated ranges and no calibration.
    Section III-E uses manufacturer intrinsics, ignores boresight and lever arm, and assumes systematic error remains acceptable.
  • domain assumption Binary segmentation masks are available each frame and can come from any detector.
    Problem setup in Sections I and III-A; the empirical mask is produced by Sobel plus morphology, not a learned model.
  • domain assumption No lens distortion; pinhole projection with manufacturer intrinsics is accurate.
    Section III-E states the lens was assumed to cause no distortions and no calibration was performed.
  • domain assumption Pixel distance between projected particles and positive pixels is a valid observation likelihood.
    Equation (8) in Section III-B; no normalization or scale is justified.
  • domain assumption Independent Gaussian diffusion in the prediction step models all target and pose uncertainty.
    Section III-B(b); variance depends on distance and an optimised constant.
  • domain assumption Simulated noise models (uniform pose noise, rectangular false positives, false negative rates) represent real-world failure modes.
    Section III-A; parameters in Table I are scenario settings, not validated against field noise statistics.
  • standard math 100,000 to 1,000,000 particles give a converged Monte Carlo approximation.
    Standard particle filter consistency, but no convergence diagnostics are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distant Object Localisation from Noisy Image Segmentation Sequences." pith.science (2026). https://pith.science/paper/HJPBHVHI

@misc{pith2026250920906,
  author       = {Pith},
  title        = {Pith review of: Distant Object Localisation from Noisy Image Segmentation Sequences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HJPBHVHI}},
  note         = {Machine review of arXiv:2509.20906}
}
read the original abstract

3D object localisation based on a sequence of camera measurements is essential for safety-critical surveillance tasks, such as drone-based wildfire monitoring. Localisation of objects detected with a camera can typically be solved with specialised sensor configurations or 3D scene reconstruction. However, in the context of distant objects or tasks limited by the amount of available computational resources, neither solution is feasible. In this paper, we show that the task can be solved with either multi-view triangulation or particle filters, with the latter also providing shape and uncertainty estimates. We studied the solutions using 3D simulation and drone-based image segmentation sequences with global navigation satellite system (GNSS) based camera pose estimates. The results suggest that combining the proposed methods with pre-existing image segmentation models and drone-carried computational resources yields a reliable system for drone-based wildfire monitoring. The proposed solutions are independent of the detection method, also enabling quick adaptation to similar tasks. Code is available at https://fgi_nls.gitlab.io/public/distant-localisation

Figures

Figures reproduced from arXiv: 2509.20906 by the authors.

Figure 1
Figure 1. We propose a hybrid approach for localising distant objects/events [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Single target simulation results. From top to bottom: Simulated camera translation from the start of the sequence, noiseless single target simulation [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Multitarget simulation results. From top to bottom: Simulated camera translation from the start of the sequence, noiseless multitarget simulation [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Empirical sequence results. From top to bottom: Time from the start of the sequence, drone-captured RGB image, segments, mean RMSE (black) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 1 linked inside Pith

  1. [1]

    Detecting wildfires on UA Vs with real-time segmentation trained by larger teacher models,

    J. Pesonen, T. Hakala, V . Karjalainen, N. Koivum ¨aki, L. Markelin, A.-M. Raita-Hakola, J. Suomalainen, I. P ¨ol¨onen, and E. Honkavaara, “Detecting wildfires on UA Vs with real-time segmentation trained by larger teacher models,” inProceedings of the Winter Conference on Applications of Computer Vision (WACV), February 2025, pp. 5166– 5176

    2025

  2. [2]

    R. I. Hartley and A. Zisserman,Multiple View Geometry in Computer Vision, 2nd ed. Cambridge University Press, ISBN: 0521540518, 2004

    2004

  3. [3]

    Structure-from-motion revisited,

    J. L. Sch ¨onberger and J.-M. Frahm, “Structure-from-motion revisited,” inConference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  4. [4]

    3D gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3D gaussian splatting for real-time radiance field rendering,”ACM Transactions on Graphics, vol. 42, no. 4, July 2023. [Online]. Available: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  5. [5]

    Object tracking: A survey,

    A. Yilmaz, O. Javed, and M. Shah, “Object tracking: A survey,”ACM computing surveys (CSUR), vol. 38, no. 4, pp. 13–es, 2006

    2006

  6. [6]

    Deep learning in video multi- object tracking: A survey,

    G. Ciaparrone, F. Luque S ´anchez, S. Tabik, L. Troiano, R. Tagliaferri, and F. Herrera, “Deep learning in video multi- object tracking: A survey,”Neurocomputing, vol. 381, pp. 61–88,

  7. [7]

    Visual object tracking: A survey,

    F. Chen, X. Wang, Y . Zhao, S. Lv, and X. Niu, “Visual object tracking: A survey,”Computer Vision and Image Understanding, vol. 222, p. 103508, 2022. [Online]. Available: https://www.sciencedirect. com/science/article/pii/S1077314222001011

    2022

  8. [8]

    A particle filter based visual object tracking: A systematic review of current trends and research challenges

    M. A. Awal, M. A. R. Refat, F. Naznin, and M. Z. Islam, “A particle filter based visual object tracking: A systematic review of current trends and research challenges.”International Journal of Advanced Computer Science & Applications, vol. 14, no. 11, 2023

    2023

Show all 35 references
  1. [9]

    Analysis based on recent deep learning approaches applied in real-time multi-object tracking: a review,

    L. Kalake, W. Wan, and L. Hou, “Analysis based on recent deep learning approaches applied in real-time multi-object tracking: a review,”IEEE Access, vol. 9, pp. 32 650–32 671, 2021

    2021

  2. [10]

    Video object segmenta- tion and tracking: A survey,

    R. Yao, G. Lin, S. Xia, J. Zhao, and Y . Zhou, “Video object segmenta- tion and tracking: A survey,”ACM Transactions on Intelligent Systems and Technology (TIST), vol. 11, no. 4, pp. 1–47, 2020

    2020

  3. [11]

    Multi-target detection and tracking from a single camera in unmanned aerial vehicles (UA Vs),

    J. Li, D. H. Ye, T. Chung, M. Kolsch, J. Wachs, and C. Bouman, “Multi-target detection and tracking from a single camera in unmanned aerial vehicles (UA Vs),” in2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016, pp. 4992–4997

    2016

  4. [12]

    Are we ready for autonomous driving? the KITTI vision benchmark suite,

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the KITTI vision benchmark suite,” inConference on Com- puter Vision and Pattern Recognition (CVPR), 2012

    2012

  5. [13]

    nuScenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuScenes: A multimodal dataset for autonomous driving,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11 621–11 631

    2020

  6. [14]

    Exploring object- centric temporal modeling for efficient multi-view 3D object detec- tion,

    S. Wang, Y . Liu, T. Wang, Y . Li, and X. Zhang, “Exploring object- centric temporal modeling for efficient multi-view 3D object detec- tion,”arXiv preprint arXiv:2303.11926, 2023

    2023 arXiv

  7. [15]

    Multi-person 3D tracking with particle filters on voxels,

    A. L ´opez, C. Canton-Ferrer, and J. R. Casas, “Multi-person 3D tracking with particle filters on voxels,” in2007 IEEE International Conference on Acoustics, Speech and Signal Processing-ICASSP’07, vol. 1. IEEE, 2007, pp. I–913

    2007

  8. [16]

    3D tracking using particle filters,

    Y . Salih and A. S. Malik, “3D tracking using particle filters,” in2011 IEEE International Instrumentation and Measurement Technology Conference. IEEE, 2011, pp. 1–4

    2011

  9. [17]

    3D tracking by catadioptric vision based on particle filters,

    M. Taiana, J. Gaspar, J. Nascimento, A. Bernardino, and P. Lima, “3D tracking by catadioptric vision based on particle filters,” inRobot Soccer World Cup. Springer, 2007, pp. 77–88

    2007

  10. [18]

    Multitarget tracking,

    B.-n. V o, M. Mallick, Y . Bar-Shalom, S. Coraluppi, R. Osborne, R. Mahler, and B.-t. V o, “Multitarget tracking,”Wiley encyclopedia of electrical and electronics engineering, no. 2015, 2015

    2015

  11. [19]

    A survey of recent advances in particle filters and remaining challenges for multitarget tracking,

    X. Wang, T. Li, S. Sun, and J. M. Corchado, “A survey of recent advances in particle filters and remaining challenges for multitarget tracking,”Sensors, vol. 17, no. 12, p. 2707, 2017

    2017

  12. [20]

    Tracking multiple objects with particle filtering,

    C. Hue, J.-P. L. Cadre, and P. P ´erez, “Tracking multiple objects with particle filtering,”IEEE Transactions on Aerospace and Electronic Systems, vol. 38, pp. 791–812, 2002. [Online]. Available: https://api.semanticscholar.org/CorpusID:14175805

    2002

  13. [21]

    Fast multiple object tracking via a hierarchical particle filter,

    C. Yang, R. Duraiswami, and L. Davis, “Fast multiple object tracking via a hierarchical particle filter,” inTenth IEEE International Confer- ence on Computer Vision (ICCV’05) V olume 1, vol. 1. IEEE, 2005, pp. 212–219

    2005

  14. [22]

    Visual object target tracking using particle filter: a survey,

    G. M. Rao and C. Satyanarayana, “Visual object target tracking using particle filter: a survey,”International Journal of Image, Graphics and Signal Processing, vol. 5, no. 6, p. 1250, 2013

    2013

  15. [23]

    Multiple particle filtering,

    P. M. Djuric, T. Lu, and M. F. Bugallo, “Multiple particle filtering,” in 2007 IEEE International Conference on Acoustics, Speech and Signal Processing-ICASSP’07, vol. 3. IEEE, 2007, pp. III–1181

    2007

  16. [24]

    Target tracking by symbiotic particle filtering,

    M. F. Bugallo and P. M. Djuri ´c, “Target tracking by symbiotic particle filtering,” in2010 IEEE Aerospace Conference. IEEE, 2010, pp. 1–7

    2010

  17. [25]

    Fusion of infor- mation in multiple particle filtering in the presence of unknown static parameters,

    X. Zhao, M. Iloska, Y . El-Laham, and M. F. Bugallo, “Fusion of infor- mation in multiple particle filtering in the presence of unknown static parameters,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  18. [26]

    Mahler,Statistical multisource-multitarget information fusion

    R. Mahler,Statistical multisource-multitarget information fusion. Artech, 2007

    2007

  19. [27]

    Particle PHD filtering for multi-target visual tracking,

    E. Maggio, E. Piccardo, C. Regazzoni, and A. Cavallaro, “Particle PHD filtering for multi-target visual tracking,” in2007 IEEE Inter- national Conference on Acoustics, Speech and Signal Processing- ICASSP’07, vol. 1. IEEE, 2007, pp. I–1101

    2007

  20. [28]

    Recent advances in multisensor multitarget tracking using random finite set,

    K. Da, T. Li, Y . Zhu, H. Fan, and Q. Fu, “Recent advances in multisensor multitarget tracking using random finite set,”Frontiers of Information Technology & Electronic Engineering, vol. 22, no. 1, pp. 5–24, 2021

    2021

  21. [29]

    Multi-camera multi- object tracking: A review of current trends and future advances,

    T. I. Amosa, P. Sebastian, L. I. Izhar, O. Ibrahim, L. S. Ayinla, A. A. Bahashwan, A. Bala, and Y . A. Samaila, “Multi-camera multi- object tracking: A review of current trends and future advances,” Neurocomputing, vol. 552, p. 126558, 2023. [Online]. Available: https: //www.s...

    2023

  22. [30]

    Multi- drone based single object tracking with agent sharing network,

    P. Zhu, J. Zheng, D. Du, L. Wen, Y . Sun, and Q. Hu, “Multi- drone based single object tracking with agent sharing network,”IEEE Transactions on Circuits and Systems for Video Technology, 2020

    2020

  23. [31]

    Robust multi-drone multi-target tracking to resolve target occlusion: A benchmark,

    Z. Liu, Y . Shang, T. Li, G. Chen, Y . Wang, Q. Hu, and P. Zhu, “Robust multi-drone multi-target tracking to resolve target occlusion: A benchmark,”IEEE Transactions on Multimedia, vol. 25, pp. 1462– 1476, 2023

    2023

  24. [32]

    A vision-based target detection, tracking, and positioning algorithm for unmanned aerial vehicle,

    X. Liu and Z. Zhang, “A vision-based target detection, tracking, and positioning algorithm for unmanned aerial vehicle,”Wireless Communications and Mobile Computing, vol. 2021, no. 1, p. 5565589, 2021

    2021

  25. [33]

    AR0234 - Arducam Wiki — docs.arducam.com,

    Arducam, “AR0234 - Arducam Wiki — docs.arducam.com,” https://docs.arducam.com/Nvidia-Jetson-Camera/Jetvariety-Camera/ AR0234/, 2022, [Accessed 04-09-2025]

    2022

  26. [34]

    Trimble APX UA V,

    Applanix, “Trimble APX UA V,” https://applanix.trimble.com/en/ products/hardware/trimble-apx-uav, [Accessed 04-09-2025]

    2025

  27. [2020]

    Available: https://www.sciencedirect.com/science/ article/pii/S0925231219315966

    [Online]. Available: https://www.sciencedirect.com/science/ article/pii/S0925231219315966

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.