Pith. sign in

REVIEW 4 major objections 5 minor 32 references

Improving trajectory continuity in drone-based crowd monitoring using a set of minimal-cost techniques and deep discriminative correlation filters

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a point-oriented SORT tracker augmented with camera-motion compensation, altitude-aware assignment, classification-based trajectory validation, and correlation filters reusing the localizer's own features reduces…

desk verdict A solid engineering contribution with a new public tracking benchmark, but the headline DDCF gains are under-verified and could partly reflect trajectory stitching rather than true identity preservation. read the letter →

arxiv 2504.20234 v1 pith:KR457BKD submitted 2025-04-28 cs.CV cs.RO

Bartosz Ptak , Marek Kraft This is my paper · ORCID
classification cs.CVcs.RO
keywords dronecrowdmonitoringmulti-objecttrackingtrajectorycontinuitydiscriminativecorrelationfilterspoint-basedlocalisationaltitude-awareassignmentidentityswitchescounting
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 argues that the main obstacle to reliable drone-based crowd monitoring is not detection accuracy but trajectory continuity: people appear as points a few pixels wide, so detector misses and false positives fragment tracks and inflate counting errors. The proposed method keeps the cheap online SORT skeleton but swaps bounding-box overlap for Euclidean point-distance assignment, adds camera-motion compensation, an altitude-dependent matching radius, a classification check before a track is confirmed, and Deep Discriminative Correlation Filters that reuse the localiser's own spatial features. On DroneCrowd and the new UP-COUNT-TRACK dataset this full combination lowers trajectory-counting error to 23% and 15%, respectively, and cuts identity switches from thousands to hundreds, outperforming an offline global-optimisation baseline. A sympathetic reader would take this as evidence that online, nearly free repair of track continuity can replace expensive offline tracking for tiny objects.

What carries the argument

The load-bearing mechanism is the reuse of the localisation network's decoder features: a 544×940×16 spatial feature map from the second-to-last layer of the head is extracted once per frame, and object-sized crops are fed into an ECO-style Deep Discriminative Correlation Filter that re-localises a confirmed trajectory whenever its detection is missing. Alongside it, association uses a circular gate with altitude-dependent radius $T_r = \max(10, \frac{100}{\text{altitude}}\cdot 10)$ pixels, and a lightweight convolutional classifier must give an average probability above 80% before a candidate track is confirmed. Together these mechanisms decide which points belong to which person and keep that decision alive through detector dropouts.

What would settle it

Take the full pipeline and disable only the DDCF re-localisation, then additionally suppress detections in a random sample of 10–30 frame windows; if the large ID-switch and Tr-nMAE improvements vanish, the filter is the cause, whereas if they survive, the other modules are carrying the gain. A second check is to crop the filter's re-localised positions and have humans or a strong detector judge whether the same person was found.

Watch

Extended reading notes

Core claim

The paper's central claim is that a point-oriented tracker can recover long, stable trajectories for sub-ten-pixel people without a new detector or offline optimisation. Replacing IoU-based association with a point-distance circle whose radius is set by flight altitude, compensating drone motion with an affine transform from sparse optical flow, validating unconfirmed tracks with a small classifier, and bridging missed detections with a Deep Discriminative Correlation Filter built on the localisation model's second-to-last-layer features reduces DroneCrowd counting error from 32% to 23% and UP-COUNT-TRACK counting error from 37% to 15%, while dropping identity switches from 6290 to 388 and from 3305 to 287. The method runs online, frame by frame, and the correlation filter adds little compute because it consumes features already computed for detection.

Load-bearing premise

The whole scheme leans on the assumption that the 16-channel feature map from the localisation head's second-to-last layer still distinguishes one few-pixel person from another person and from background during missed-detection intervals; if those features are not identity-discriminative at that scale, the correlation filter can drift and the ID-switch and counting gains would reflect prolonged but wrong trajectories rather than true tracking.

Editorial extensions

If this is right

  • Trajectory-counting error falls to 23% on DroneCrowd and 15% on UP-COUNT-TRACK, so counting people by unique trajectories becomes a practical route to crowd-size estimates.
  • Identity switches drop from 6290 to 388 on DroneCrowd and from 3305 to 287 on UP-COUNT-TRACK, which would make per-person movement and behaviour analysis feasible instead of just counting.
  • Because the enhanced online tracker beats the offline greedy global-optimisation method on UP-COUNT-TRACK and is competitive on DroneCrowd, the paper implies offline processing is not required for good trajectory continuity.
  • The altitude-dependent assignment radius lets a single configuration handle flights from roughly 29 m to 100 m above ground, removing a tuning step for different drone heights.
  • Reusing localiser features avoids a second feature-extraction network, but the paper notes the overall pipeline remains computationally demanding and is not yet real-time on board a drone.

Reading between the lines

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

  • The reported ID-switch collapse may be partly a consequence of artificially prolonging trajectories through detection gaps; a direct way to test this is to compare tracks produced with and without the DDCF module on sequences where detector outputs are deliberately suppressed.
  • If the 16-channel features carry enough identity information for few-pixel objects, the same feature-reuse pattern could extend to other point-based dense prediction tasks, such as vehicle or animal counting, that output spatial feature maps from their heads.
  • The altitude rule uses only a scalar radius; incorporating full camera pose from GPS, IMU, and gimbal angles could generalise the assignment gate to non-horizontal cameras and curved flight paths.
  • The order-of-magnitude ID-switch reduction suggests many identity switches in drone datasets are artifacts of detection gaps rather than association ambiguities; detector training might benefit from a temporal continuity loss that rewards stable features across frames.
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

4 major / 5 minor

Summary. The paper proposes an online, point-oriented multi-object tracker for drone-based crowd monitoring, built on SORT with a point-distance assignment metric. It adds three minimal-cost techniques: camera motion compensation (CMC), altitude-aware dynamic assignment thresholds, and a classification-based trajectory validation step. It further integrates Deep Discriminative Correlation Filters (DDCF) that reuse spatial feature maps from the Dot Localisation network, aiming to maintain trajectory continuity during missed detections. The method is evaluated on the newly released UP-COUNT-TRACK dataset and on DroneCrowd, reporting reduced identity switches (ID-SW) and trajectory counting errors (Tr-nMAE) relative to a baseline SORT variant and to the offline GOG algorithm.

Significance. If the DDCF module genuinely preserves object identity during missed-detection gaps, the paper offers a practical, computationally efficient online recipe for tracking sub-ten-pixel people in drone footage, with a valuable new benchmark dataset (UP-COUNT-TRACK) and a zero-waste reuse of localisation features. The incremental ablation design is commendable: Tables 2 and 3 show that each added component either improves or holds the reported metrics. However, the central claimed breakthrough—the drastic ID-SW reduction attributed to DDCF—is not supported by direct evidence of per-object re-identification fidelity, and the comparison set lacks established online trackers. The dataset contribution alone is significant for the community.

major comments (4)
  1. [Section 4.4-4.5 and Tables 2-3] The DDCF module is credited with the dominant improvement: ID-SW drops from 2943 to 287 on UP-COUNT-TRACK and from 6645 to 388 on DroneCrowd, while HOTA remains essentially flat (0.63 to 0.63; 0.52 to 0.54). The paper does not demonstrate that the correlation-filter-based re-detections recover the same physical person; for objects of only a few pixels, the 16-channel feature map from the localisation head may not be identity-discriminative, and the filter could lock onto a neighbor or background. Without a per-track re-identification accuracy, a failure analysis, a comparison against cheaper gap-filling (e.g., Kalman prediction alone), or a HOTA decomposition showing association gains, the ID-SW reduction is equally consistent with trajectory stitching that artificially prolongs tracks. This is load-bearing: the central claim of improved trajectory continuity hinges on the DDCF's identity-preserving behaviour.
  2. [Section 5.3] The comparison includes only the paper's own baseline with incremental components and the offline GOG algorithm. The abstract and Section 1 claim the method 'outperform[s] baseline online trackers', but no established online multi-object trackers (e.g., DeepSORT, ByteTrack, BoT-SORT, or a standard point-based SORT variant) are evaluated. Since SORT is acknowledged as the base, a comparison against a well-tuned SORT/DeepSORT would be necessary to substantiate the practical improvement claim.
  3. [Eq. (1) and Section 4.3.2] The altitude-aware threshold for DroneCrowd relies on assuming 100 m for 'small' objects and 50 m for 'large' objects because the dataset lacks altitude metadata. The paper does not justify these specific values or report sensitivity to them. Given that the threshold directly controls assignment radii and thus ID-SW, a sensitivity analysis (e.g., varying the assumed altitudes or the constant in Eq. (1)) is needed to show the reported gains are not artifacts of tuned assumptions.
  4. [Section 4.3.3 and Section 5.5] The classification-based validation uses an 80% average-probability threshold stated as 'determined experimentally', but no ablation on this threshold is reported, and the training/evaluation protocol for the classifier is underspecified: is the classifier trained on UP-COUNT-TRACK training split and then evaluated on DroneCrowd, or trained separately per dataset? The accuracy figures in Section 5.5 (0.987 and 0.939) do not indicate whether the classifier generalises across domains or whether the threshold was tuned on the test sets.
minor comments (5)
  1. [Abstract] There is a stray spacing issue: 'drone -based' should be 'drone-based'.
  2. [Tables 2 and 3] The numeric formatting of '49 .13± 117 .22' and similar entries inserts spaces before decimal points; these should be corrected throughout.
  3. [Section 4.4.1] The spatial feature map dimensions 544×940×16 are mentioned without stating the input image resolution to which they correspond; adding this context would help readers judge the feature stride relative to object size.
  4. [Section 5.2] The definitions of Tr-MAE and Tr-nMAE use n for the number of sequences, but it is not explicitly stated that y_i and ŷ_i are the ground-truth and estimated numbers of unique trajectories per sequence; please make this explicit, as the metric is central to the counting claims.
  5. [References] The BoT-SORT reference is incomplete: it lists 'Aharon, N., Orfaig, R., Bobrovsky, B., .' with no year; the arXiv identifier is given but the citation entry should be completed.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the tracking gains are measured against external benchmarks; DDCF is a pipeline component, not a fitted predictor, and self-citations are non-load-bearing.

full rationale

The paper's derivation chain is not circular. Detections come from the authors' previously published Dot Localisation method, which is used as an external input and benchmarked (L-AP@10 on DroneCrowd and UP-COUNT), not as a quantity derived from the tracking metrics. The tracking pipeline applies SORT with point-distance assignment, camera motion compensation, an altitude-dependent threshold (Eq. 1), an early-classification trigger (Eq. 2), an experimentally determined 80% confirmation threshold, and an ECO-style DDCF that reuses dense feature maps from the localisation head. None of the reported metrics (HOTA, T-mAP, ID-SW, Tr-MAE, Tr-nMAE) is algebraically identical to any fitted parameter or to the DDCF feature maps; the reductions in ID-SW and Tr-nMAE are empirical measurements on DroneCrowd and the newly constructed UP-COUNT-TRACK dataset. Eq. (1) is a heuristic mapping from altitude to pixel radius, Eq. (2) is a trajectory-age test, and the 80% threshold is stated as experimentally determined; these are engineering choices, not identities. The DDCF could be under-validated as a per-object re-identification mechanism, but that is a measurement-interpretation or correctness concern, not circular reasoning. The self-citations to Ptak and Kraft (2025) provide the localisation backbone and dataset, but the central tracking claim is independently evaluated against external metrics and an external benchmark, so no load-bearing reduction to the paper's own inputs occurs.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method contributes a combination of known engineering components; no new theoretical axioms or physical entities are introduced. Most free parameters are heuristic thresholds set by hand or experimentally without sensitivity analysis. The most novel assumption is that localisation features reused for DDCF retain per-identity discriminative power, flagged as ad_hoc_to_paper.

free parameters (7)
  • Assignment radius default = 10 px
    Default matching radius equals half the 20 px object size from DroneCrowd; used when altitude metadata is absent or threshold falls below 10 px (Section 4.2, Eq. 1).
  • Dynamic threshold altitude factor = Tr = max(10, (100 / altitude) * 10)
    Heuristic mapping from altitude to pixel threshold in Eq. 1; no derivation or sensitivity analysis.
  • Minimum trajectory length (Nthresh) = 30 frames
    Trajectories must persist roughly 1 second before confirmation; value chosen for robustness, no ablation reported.
  • Maximum trajectory age = 60 frames
    Unmatched trajectories kept for roughly 2 seconds before removal; chosen, no ablation reported.
  • Classification validation threshold = 80% average probability
    Trajectory declared valid only if classified localisation surroundings average more than 80% person; determined experimentally (Section 4.3.3).
  • Confirmation trigger offset = Nthresh - 3
    Classification begins three frames before nominal confirmation; ad hoc offset in Eq. 2.
  • DroneCrowd assumed altitudes = 100 m for small-object sequences, 50 m for large-object sequences
    DroneCrowd has no altitude metadata; altitudes are assumed from the small/large object-size split, feeding Eq. 1 (Section 4.3.2).
assumptions (5)
  • standard math The Hungarian algorithm yields the optimal one-to-one assignment for the distance cost matrix.
    Used in Section 4.5 for detection-to-trajectory matching; a standard combinatorial result, not under dispute.
  • domain assumption SORT's Kalman filter approximates each person's motion as linear and Gaussian over short intervals.
    Section 4.2 inherits SORT's motion model; drone ego-motion and abrupt stops can violate it, but CMC is intended to absorb some of this.
  • domain assumption Inter-frame camera motion is representable by a single affine transformation estimated from sparse Lucas-Kanade corners.
    Adopted from BoT-SORT (Section 4.3.1); parallax and independent object motion can break a global affine model, and no robustness analysis is provided.
  • ad hoc to paper The 16-channel spatial feature map from the localisation head is discriminative enough for DDCF re-detection of individual tiny people.
    Assumed in Sections 4.4.1 and 4.5; the paper does not measure feature discriminability or per-object re-detection success, only the final tracking metrics.
  • domain assumption The trajectory-validation classifier trained on known object positions and random negatives generalizes to test sequences without label leakage.
    Section 4.3.3 describes training data construction but not the train/test split; if ground-truth from test sequences was used, classification accuracies in Section 5.5 would be inflated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving trajectory continuity in drone-based crowd monitoring using a set of minimal-cost techniques and deep discriminative correlation filters." pith.science (2026). https://pith.science/paper/KR457BKD

@misc{pith2026250420234,
  author       = {Pith},
  title        = {Pith review of: Improving trajectory continuity in drone-based crowd monitoring using a set of minimal-cost techniques and deep discriminative correlation filters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KR457BKD}},
  note         = {Machine review of arXiv:2504.20234}
}
read the original abstract

Drone-based crowd monitoring is the key technology for applications in surveillance, public safety, and event management. However, maintaining tracking continuity and consistency remains a significant challenge. Traditional detection-assignment tracking methods struggle with false positives, false negatives, and frequent identity switches, leading to degraded counting accuracy and making in-depth analysis impossible. This paper introduces a point-oriented online tracking algorithm that improves trajectory continuity and counting reliability in drone-based crowd monitoring. Our method builds on the Simple Online and Real-time Tracking (SORT) framework, replacing the original bounding-box assignment with a point-distance metric. The algorithm is enhanced with three cost-effective techniques: camera motion compensation, altitude-aware assignment, and classification-based trajectory validation. Further, Deep Discriminative Correlation Filters (DDCF) that re-use spatial feature maps from localisation algorithms for increased computational efficiency through neural network resource sharing are integrated to refine object tracking by reducing noise and handling missed detections. The proposed method is evaluated on the DroneCrowd and newly shared UP-COUNT-TRACK datasets, demonstrating substantial improvements in tracking metrics, reducing counting errors to 23% and 15%, respectively. The results also indicate a significant reduction of identity switches while maintaining high tracking accuracy, outperforming baseline online trackers and even an offline greedy optimisation method.

Figures

Figures reproduced from arXiv: 2504.20234 by the authors.

Figure 1
Figure 1. An example frame demonstrates the task of localising and tracking tiny objects. Marked trajectories represent [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Example frames with annotated trajectories illustrate the dataset’s diverse recording environments, varying [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Architecture of a simple convolutional network designed to classify if a region of interest contains a person. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Usage of Deep Discriminative Correlation Filters in drone-based people tracking for sample images. The [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Sixteen deep features extracted in the neighbourhood of a person. Values are normalised to a zero-one range [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Overview of the proposed tracking pipeline, integrating point-based object localisation, spatial feature maps [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The visual comparison of ground-truth trajectories (green) and estimated results (red) for both datasets. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Statistical analysis of counting trajectory error and three sequence characteristics: sequence length, number of [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Confusion matrices of the classification algorithm for the UP-COUNT-TRACK and DroneCrowd datasets. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Example screenshots from a visualisation application that combines people’s trajectories with drone sensors [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 27 canonical work pages

  1. [1]

    , author Orfaig, R

    author Aharon, N. , author Orfaig, R. , author Bobrovsky, B. , . title BoT-SORT : Robust associations multi-pedestrian tracking . journal arXiv preprint arXiv:2206.14651

    arXiv

  2. [2]

    , author Nishimura, K

    author Asanomi, T. , author Nishimura, K. , author Bise, R. , year 2023 . title Multi-frame attention with feature-level warping for drone crowd tracking , in: booktitle Winter Conf. on Applications of Computer Vision , pp. pages 1664--1673

    work page 2023

  3. [3]

    , author Stiefelhagen, R

    author Bernardin, K. , author Stiefelhagen, R. , year 2008 . title Evaluating multiple object tracking performance: the clear MOT metrics . journal EURASIP Journal on Image and Video Processing volume 2008 , pages 1--10

    work page 2008

  4. [4]

    , author Ge, Z

    author Bewley, A. , author Ge, Z. , author Ott, L. , author Ramos, F. , author Upcroft, B. , year 2016 . title Simple online and realtime tracking , in: booktitle 2016 IEEE international conference on image processing (ICIP) , organization IEEE . pp. pages 3464--3468

    work page 2016

  5. [5]

    , et al., year 2001

    author Bouguet, J.Y. , et al., year 2001 . title Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm . journal Intel corporation volume 5 , pages 4

    work page 2001

  6. [6]

    , year 2014

    author Chatfield, K. , year 2014 . title Return of the devil in the details: Delving deep into convolutional nets . journal arXiv preprint arXiv:1405.3531

    arXiv 2014

  7. [7]

    , author Bhat, G

    author Danelljan, M. , author Bhat, G. , author Shahbaz Khan, F. , author Felsberg, M. , year 2017 . title ECO : Efficient convolution operators for tracking , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 6638--6646

    work page 2017

  8. [8]

    , author Hager, G

    author Danelljan, M. , author Hager, G. , author Shahbaz Khan, F. , author Felsberg, M. , year 2015 . title Convolutional features for correlation filter based visual tracking , in: booktitle Proceedings of the IEEE international conference on computer vision workshops , pp. pages 58--66

    work page 2015

Show all 32 references
  1. [9]

    , author Robinson, A

    author Danelljan, M. , author Robinson, A. , author Shahbaz Khan, F. , author Felsberg, M. , year 2016 . title Beyond correlation filters: Learning continuous convolution operators for visual tracking , in: booktitle Computer Vision--ECCV 2016: 14th European Conference, Amster...

    2016

  2. [10]

    , author Shahbaz Khan, F

    author Danelljan, M. , author Shahbaz Khan, F. , author Felsberg, M. , author Van de Weijer, J. , year 2014 . title Adaptive color attributes for real-time visual tracking , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages...

    2014

  3. [11]

    , author Khaleel, M

    author Emimi, M. , author Khaleel, M. , author Alkrash, A. , year 2023 . title The current opportunities and challenges in drone technology . journal Int. J. Electr. Eng. and Sustain. , pages 74--89

    2023

  4. [12]

    , author Girshick, R.B

    author Felzenszwalb, P.F. , author Girshick, R.B. , author McAllester, D. , author Ramanan, D. , year 2009 . title Object detection with discriminatively trained part-based models . journal IEEE transactions on pattern analysis and machine intelligence volume 32 , pages 1627--1645

    2009

  5. [13]

    , author Sim, T

    author Galoogahi, H.K. , author Sim, T. , author Lucey, S. , year 2013 . title Multi-channel correlation filters , in: booktitle Proceedings of the IEEE international conference on computer vision , pp. pages 3072--3079

    2013

  6. [14]

    , author Caseiro, R

    author Henriques, J.F. , author Caseiro, R. , author Martins, P. , author Batista, J. , year 2014 . title High-speed tracking with kernelized correlation filters . journal IEEE transactions on pattern analysis and machine intelligence volume 37 , pages 583--596

    2014

  7. [15]

    , year 2014

    author Kingma, D.P. , year 2014 . title Adam: A method for stochastic optimization . journal arXiv preprint arXiv:1412.6980

    2014 arXiv

  8. [16]

    , author Matas, J

    author Kristan, M. , author Matas, J. , author Leonardis, A. , author Felsberg, M. , author Cehovin, L. , author Fernandez, G. , author Vojir, T. , author Hager, G. , author Nebehay, G. , author Pflugfelder, R. , year 2015 . title The visual object tracking VOT2015 challenge r...

    2015

  9. [17]

    , author Huang, C

    author Li, Y. , author Huang, C. , author Nevatia, R. , year 2009 . title Learning to associate: Hybridboosted multi-target tracker for crowded scene , in: booktitle 2009 IEEE Conference on Computer Vision and Pattern Recognition , pp. pages 2953--2960 . :10.1109/CVPR.2009.5206735

    2009

  10. [18]

    , author Zhu, J

    author Li, Y. , author Zhu, J. , year 2015 . title A scale adaptive kernel correlation filter tracker with feature integration , in: booktitle Computer Vision-ECCV 2014 Workshops: Zurich, Switzerland, September 6-7 and 12, 2014, Proceedings, Part II 13 , organization Springer ...

    2015

  11. [19]

    , author Osep, A

    author Luiten, J. , author Osep, A. , author Dendorfer, P. , author Torr, P. , author Geiger, A. , author Leal-Taix \'e , L. , author Leibe, B. , year 2021 . title HOTA : A higher order metric for evaluating multi-object tracking . journal International journal of computer vis...

    2021

  12. [20]

    , author Vojir, T

    author Lukezic, A. , author Vojir, T. , author ˇCehovin Zajc, L. , author Matas, J. , author Kristan, M. , year 2017 . title Discriminative correlation filter with channel and spatial reliability , in: booktitle Proceedings of the IEEE conference on computer vision and pattern...

    2017

  13. [21]

    , author Huang, J.B

    author Ma, C. , author Huang, J.B. , author Yang, X. , author Yang, M.H. , year 2015 . title Hierarchical convolutional features for visual tracking , in: booktitle Proceedings of the IEEE international conference on computer vision , pp. pages 3074--3082

    2015

  14. [22]

    , author Gen \'e -Mola, J

    author Miranda, J.C. , author Gen \'e -Mola, J. , author Zude-Sasse, M. , author Tsoulias, N. , author Escol \`a , A. , author Arn \'o , J. , author Rosell-Polo, J.R. , author Sanz-Cortiella, R. , author Mart \' nez-Casasnovas, J.A. , author Gregorio, E. , year 2023 . title Fr...

    2023

  15. [23]

    , author Nezamabadi-Pour, H

    author Mirzaei, B. , author Nezamabadi-Pour, H. , author Raoof, A. , author Derakhshani, R. , year 2023 . title Small object detection and tracking: a comprehensive review . journal Sensors volume 23 , pages 6887

    2023

  16. [24]

    , author Ramanan, D

    author Pirsiavash, H. , author Ramanan, D. , author Fowlkes, C.C. , year 2011 . title Globally-optimal greedy algorithms for tracking a variable number of objects , in: booktitle CVPR 2011 , organization IEEE . pp. pages 1201--1208

    2011

  17. [25]

    , author Kraft, M

    author Ptak, B. , author Kraft, M. , year 2025 . title Enhancing people localisation in drone imagery for better crowd management by utilising every pixel in high-resolution images . https://arxiv.org/abs/2502.04014, http://arxiv.org/abs/2502.04014 arXiv:2502.04014

    2025 arXiv

  18. [26]

    , author Twardowski, B

    author Trzcinski, T. , author Twardowski, B. , author Zieli \'n ski, B. , author Adamczewski, K. , author W \'o jcik, B. , year 2024 . title Zero-waste machine learning , in: booktitle ECAI 2024 . publisher IOS Press , pp. pages 43--49

    2024

  19. [27]

    , author Schmid, C

    author Van De Weijer, J. , author Schmid, C. , author Verbeek, J. , author Larlus, D. , year 2009 . title Learning color names for real-world applications . journal IEEE Transactions on Image Processing volume 18 , pages 1512--1523

    2009

  20. [28]

    , author Liu, T

    author Wang, L. , author Liu, T. , author Wang, G. , author Chan, K.L. , author Yang, Q. , year 2015 . title Video tracking using learned hierarchical features . journal IEEE Transactions on Image Processing volume 24 , pages 1424--1435

    2015

  21. [29]

    , author Yeung, D.Y

    author Wang, N. , author Yeung, D.Y. , year 2013 . title Learning a deep compact image representation for visual tracking . journal Advances in neural information processing systems volume 26

    2013

  22. [30]

    , author Du, D

    author Wen, L. , author Du, D. , author Zhu, P. , author Hu, Q. , author Wang, Q. , author Bo, L. , author Lyu, S. , year 2021 . title Detection, tracking, and counting meets drones in crowds: A benchmark , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Visio...

    2021

  23. [31]

    , author Ptak, B

    author Wilinski, M. , author Ptak, B. , author Kraft, M. , year 2024 . title Elevating point-based object detection in UAVs : A deep learning method with altitude fusion , in: booktitle Progress in Polish Artificial Intelligence Research 5: Proceedings of the 5th Polish Confer...

    2024

  24. [32]

    , author Lim, J

    author Wu, Y. , author Lim, J. , author Yang, M.H. , year 2013 . title Online object tracking: A benchmark , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 2411--2418

    2013

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.