Cited by 3,371Video Surveillance and Tracking MethodsAutonomous Vehicle Technology and SafetyAdvanced Neural Network Applications

The DOI matches its Crossref record; 97 of the 118 references it cites were found in a registry or an index; none of the 11 sources its claims cite is checked for support yet.

Object Tracking Benchmark

Wu, Yi, Lim, Jongwoo, Yang, Ming-Hsuan·2015·IEEE Transactions on Pattern Analysis and Machine Intelligence

DOI 10.1109/tpami.2014.2388226

Open this paper in CoChat — free account

Abstract

Object tracking has been one of the most important and active research areas in the field of computer vision. A large number of tracking algorithms have been proposed in recent years with demonstrated success. However, the set of sequences used for evaluation is often not sufficient or is sometimes biased for certain types of algorithms. Many datasets do not have common ground-truth object positions or extents, and this makes comparisons among the reported quantitative results difficult. In addition, the initial conditions or parameters of the evaluated tracking algorithms are not the same, and thus, the quantitative results reported in literature are incomparable or sometimes contradictory. To address these issues, we carry out an extensive evaluation of the state-of-the-art online object-tracking algorithms with various evaluation criteria to understand how these methods perform within the same framework. In this work, we first construct a large dataset with ground-truth object positions and extents for tracking and introduce the sequence attributes for the performance analysis. Second, we integrate most of the publicly available trackers into one code library with uniform input and output formats to facilitate large-scale performance evaluation. Third, we extensively evaluate the performance of 31 algorithms on 100 sequences with different initialization settings. By analyzing the quantitative results, we identify effective approaches for robust tracking and provide potential future research directions in this field.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper presents a large-scale benchmark for online single-object tracking, with 50 annotated sequences, a code library of 29 trackers, and one-pass, temporal, and spatial robustness evaluations. It uses precision and success plots plus sequence attributes to identify effective tracking components and future research directions.

Key findings

The benchmark includes 50 fully annotated sequences, 11 attributes, 29 tracking algorithms, and more than 660,000 bounding box outputs; the OCC subset contains 29 sequences.
In success plots, top-ranked SCM in OPE outperforms Struck by 2.6%, but SCM is 1.9% below Struck in SRE; OPE is described as not the best performance indicator because it is one trial of SRE or TRE.
Sparse-representation trackers perform well in SRE and TRE; local sparse representations (SCM, ASLA, LSK) outperform holistic sparse templates (MTT, L1APG). On OCC, Struck, SCM, TLD, LSK, and ASLA outperform others; on SV, ASLA, SCM, and Struck perform best; for fast motion, dense-sampling trackers such as Struck, TLD, and CXT perform much better than others.
Performance often decreases significantly when initialization scale is large (e.g., ×1.2); Struck is less sensitive to scale variation than other well-performing methods, and CSK has the highest speed among the top 10 due to its circulant structure.

Methods

Built a tracking dataset of 50 fully annotated sequences with 11 attributes and a code library integrating 29 publicly available trackers with uniform input/output formats, plus VIVID testbed methods.
Used precision plots based on center location error, with representative score at threshold = 20 pixels, and success plots based on bounding box overlap, ranked by area under curve (AUC).
Evaluated with one-pass evaluation (OPE), temporal robustness evaluation (TRE; each sequence partitioned into 20 segments), and spatial robustness evaluation (SRE; 12 runs per sequence using 8 spatial shifts and 4 scale variations, with shift amount 10% of target size and scale ratios 0.8, 0.9, 1.1, 1.2). OPE tested each tracker on more than 29,000 frames, SRE generated more than 350,000 bounding box results, and TRE performed on around 310,000 frames.

Limitations

Stated: due to space limitations, the main paper presents only the top 10 trackers and only the OCC, SV, and FM attribute subsets; complete plots are in the supplementary material.
Stated/evident: evaluation is restricted to trackers whose original source or binary codes are publicly available and uses default parameters from the source code; the work focuses on online single-target tracking, and the paper notes there is no single tracking approach that can successfully handle all scenarios.
Evident: conclusions are tied to the 50-sequence dataset and its attribute thresholds, such as SV with \(t_s=2\), FM with \(t_m=20\), and LR with \(t_r=400\).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/tpami.2014.2388226

Cites

97 of 97 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 11 cited sources checked · 0 supported · 0 flagged

The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Additionally each sequence in the dataset is annotated with attributes that often affect tracking performance, such as occlusion, fast motion, and illumination variation.”
- Also stated in: § Abstract · p. 1 · plain
  “The test image sequences are annotated with different attributes for performance evaluation and analysis.”
Load-bearingStrength:plain· not yet calibratedCausal:correlational· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“To facilitate fair performance evaluation, we have collected and annotated most commonly used tracking sequences.”
Load-bearingStrength:plain· not yet calibratedCausal:direct· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
- Rests on an assumption: Most sequences do not have the ground truth annotations, and the quantitative evaluation results may be generated with different initial conditions.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“We build a tracking dataset with 50 fully annotated sequences to facilitate tracking evaluation.”
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
- Rests on an assumption: Most sequences do not have the ground truth annotations, and the quantitative evaluation results may be generated with different initial conditions.

Show all 43 claims

It may state two separate claims as one. 12
Probably not a substantial claim. 26

The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In this work, we build a code library that includes most publicly available trackers and a test dataset with ground-truth annotations to facilitate the evaluation task.”
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
It may state two separate claims as one.
- Rests on an assumption: The input and output formats of most trackers are different and thus it is inconvenient for large scale performance evaluation.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The initial bounding boxes for tracking are sampled spatially and temporally to evaluate the robustness and characteristics of trackers.”
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
Probably not a substantial claim.
- Rests on an assumption: The robustness to initialization is a well-known problem in the field, but it is seldom addressed in the literature.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“the target objects are usually humans or cars of small size in these surveillance sequences, and the background is usually static.”
What the sources say
- Not matched
  Couldn’t be matched to a paper in Crossref, so it can’t be checked.
  cites [14] [14] R. Collins, X. Zhou, and S. K. Teh. An Open Source Tracking Testbed and Evaluation Web Site. In PETS , 2005. -
- Not matched
  Couldn’t be matched to a paper in Crossref, so it can’t be checked.
  cites [21] [21] R. B. Fisher. The PETS04 Surveillance Ground-Truth Data Sets. In PETS , 2004. -
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, we propose two ways to analyze a tracker's robustness to initialization, by perturbing the initialization temporally (i.e., start at different frames) and spatially (i.e., start by different bounding boxes).”
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
Probably not a substantial claim.
- Rests on an assumption: The robustness to initialization is a well-known problem in the field, but it is seldom addressed in the literature.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“For fair and comprehensive performance evaluation, we propose to perturb the initial state spatially and temporally from the ground-truth target locations.”
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
Probably not a substantial claim.
- Rests on an assumption: The robustness to initialization is a well-known problem in the field, but it is seldom addressed in the literature.
In the paper
“However, when the tracker loses the target, the output location can be random and the average error value may not measure the tracking performance correctly [6].”
What the sources say
- Not checked yetcites [6] [6] B. Babenko, M.-H. Yang, and S. Belongie. Robust Object Tracking with Online Multiple Instance Learning. PAMI , 33(7):1619-1632, 2011. - source
Load-bearingStrength:hedged· not yet calibratedCausal:conditional· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“To the best of our knowledge, this is the first comprehensive work to address and analyze the initialization problem of object tracking.”
Strength:hedged· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✗ Based: fails
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
- Rests on an assumption: The robustness to initialization is a well-known problem in the field, but it is seldom addressed in the literature.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The initialization errors tend to cause trackers to update with imprecise appearance information, thereby causing gradual drifts.”
Load-bearingStrength:plain· not yet calibratedCausal:direct· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: fails
Probably not a substantial claim. The paper may give no reason why it holds.
- Rests on an assumption: Trackers are often initialized by an object detector, which is likely to introduce initialization errors in terms of position and scale.
None of the sources cited for this claim have been checked yet.
In the paper
“Recently the precision plot [6, 27] has been adopted to measure the overall tracking performance.”
What the sources say
- Not checked yetcites [6] [6] B. Babenko, M.-H. Yang, and S. Belongie. Robust Object Tracking with Online Multiple Instance Learning. PAMI , 33(7):1619-1632, 2011. - source
- Not checked yetcites [27] [27] F. Henriques, R. Caseiro, P. Martins, and J. Batista. Exploiting the Circulant Structure of Tracking-by-Detection with Kernels. In ECCV , 2012. - source
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
Probably not a substantial claim.
None of the sources cited for this claim have been checked yet.
In the paper
“Although object tracking has been studied for several decades, and much progress has been made in recent years [28, 16, 47, 5, 40, 26, 19], it remains a very challenging problem.”
What the sources say
- Not checked yetcites [28] [28] M. Isard and A. Blake. CONDENSATION-Conditional Density Propagation for Visual Tracking. IJCV , 29(1):5-28, 1998. - source
- Not checked yetcites [16] [16] D. Comaniciu, V. Ramesh, and P. Meer. Kernel-Based Object Tracking. PAMI , 25(5):564-577, 2003. - source
- Not checked yetcites [47] [47] D. Ross, J. Lim, R.-S. Lin, and M.-H. Yang. Incremental Learning for Robust Visual Tracking. IJCV , 77(1):125-141, 2008. - source
Load-bearingStrength:plain· not yet calibrated
Tests✓ Asserted: passes✗ Checkable: fails✓ Load-bearing: passes✓ Based: passes
Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.

33 of 33

The paper relies on this without support.
The robustness to initialization is a well-known problem in the field, but it is seldom addressed in the literature.
“While the robustness to initialization is a well-known problem in the field, it is seldom addressed in the literature.”
The paper relies on this without support.
Most sequences do not have the ground truth annotations, and the quantitative evaluation results may be generated with different initial conditions.
“However, most sequences do not have the ground truth annotations, and the quantitative evaluation results may be generated with different initial conditions.”
The paper relies on this without support.
The input and output formats of most trackers are different and thus it is inconvenient for large scale performance evaluation.
“However, the input and output formats of most trackers are different and thus it is inconvenient for large scale performance evaluation.”
The paper relies on this without support.
Trackers are often initialized by an object detector, which is likely to introduce initialization errors in terms of position and scale.
“as a tracker is often initialized by an object detector, which is likely to introduce initialization errors in terms of position and scale.”
The paper relies on this without support.
An object detector may be used to re-initialize a tracker at different time instances.
“In addition, an object detector may be used to re-initialize a tracker at different time instances.”
The paper relies on this without support.
TLD performs well in long sequences with a redetection module.
“This is because TLD performs well in long sequences with a redetection module while there are numerous short segments in TRE.”
The paper relies on this without support.
Struck only estimates the location of target and does not handle scale variation.
“This is because Struck only estimates the location of target and does not handle scale variation.”
The paper relies on this without support.
Sparse representations are used in SCM, ASLA, LSK, MTT and L1APG.
“Sparse representations are used in SCM, ASLA, LSK, MTT and L1APG.”
The paper relies on this without support.
The VTD and VTS methods adopt mixture models to improve the tracking performance.
“The VTD and VTS methods adopt mixture models to improve the tracking performance.”
The paper relies on this without support.
Trackers are often sensitive to initialization variations.
“It has been known that trackers are often sensitive to initialization variations.”
The paper relies on this without support.
The templates have to be warped to fit the size of the usually smaller canonical template.
“the templates have to be warped to fit the size of the usually smaller canonical template”

11 of 11

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Histograms of Oriented Gradients for Human Detection
Dalal, N., & Triggs, B. (2005) · 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) · doi:10.1109/cvpr.2005.177 · Cited by 33,584
The pascal visual object classes (VOC) challenge
Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010) · International Journal of Computer Vision · doi:10.1007/s11263-009-0275-4 · Cited by 19,599
Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
Ojala, T., Pietikainen, M., & Maenpaa, T. (2002) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2002.1017623 · Cited by 15,398
Robust Real-Time Face Detection
Viola, P., & Jones, M. J. (2004) · International Journal of Computer Vision · doi:10.1023/b:visi.0000013087.49260.fb · Cited by 14,097
Untitled work
(1981) · Cited by 11,963
High-Speed Tracking with Kernelized Correlation Filters
Henriques, J. F., Caseiro, R., Martins, P., & Batista, J. (2015) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2014.2345390 · Cited by 5,753
GrabCut: Interactive foreground extraction using iterated graph cuts
Rother, C., Kolmogorov, V., & Blake, A. (2004) · ACM Transactions on Graphics · doi:10.1145/1015706.1015720 · Cited by 5,703
Object tracking
Yılmaz, A., Javed, O., & Shah, M. A. (2006) · ACM Computing Surveys · doi:10.1145/1177352.1177355 · Cited by 4,902
CONDENSATION—Conditional Density Propagation for Visual Tracking
Isard, M., & Blake, A. (1998) · International Journal of Computer Vision · doi:10.1023/a:1008078328650 · Cited by 4,886
Kernel-based object tracking
Comaniciu, D., Ramesh, V., & Meer, P. (2003) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2003.1195991 · Cited by 4,560

Flagged first, then by citations

Show all 118 references

21 of the 118 references couldn’t be matched to a known paper.

Research works printed but not linked: 5

vojir, 0, Robustifying the flock of trackers, Proc Comput Vis Winter Workshop
fisher, 0, The PETS04 surveillance ground-truth data sets, Proc IEEE Int Workshop Perform Eval Tracking Surveillance, 1
ferryman, 0, PETS 2009: Dataset and challenge, Proc IEEE Int Workshop Perform Eval Tracking Surveillance, 1
mei, 0, Robust visual tracking using l1 minimization, Proc 12th IEEE Int Conf Comput Vis, 1436
collins, 0, An open source tracking testbed and evaluation web site, Proc IEEE Int Workshop Perform Eval Tracking Surveillance

Sorted by citations

Items 1,001–1,002 of 1,002

Embedding Symbolic Temporal Knowledge into Deep Sequential Models
Xie, Y., Zhou, F., & Soh, H. (2021) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra48506.2021.9561952 · Cited by 16
Multiple Cues-Based Robust Visual Object Tracking Method
Khan, B., Jalil, A., Ali, A., Alkhaledi, K., Mehmood, K., Cheema, K. M., Murad, M., Tariq, H., & El-Sherbeeny, A. M. (2022) · Electronics · doi:10.3390/electronics11030345 · Cited by 5
“Twenty-three videos were selected from three standard datasets, i.e., OTB-50 [8], OTB-100 [9], and Temple Colour-128 [10].”
Influential

Authors: Wu, Y., Lim, J., & Yang, M.-H.
Affiliations: Nanjing University of Information Science and Technology (CN)
Hanyang University (KR)
University of California, Merced (US)
ORCID: Wu, Yi
Lim, Jongwoo
Yang, Ming-Hsuan
Venue: IEEE Transactions on Pattern Analysis and Machine Intelligence
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Published: 2015-09-01
Volume: 37
Issue: 9
Pages: 1834-1848
ISSN: 0162-8828 (ISSN-L)
2160-9292
DOI: 10.1109/tpami.2014.2388226
OpenAlex: W2158592639
Semantic Scholar: c4c45661501c16064eead6e5d37dcb80d41c7a78
Cited by: 3,371
References: 118
Citations by year: 2026: 102 · 2025: 174 · 2024: 209 · 2023: 295 · 2022: 323 · 2021: 442 · 2020: 483 · 2019: 539 · 2018: 429 · 2017: 235 · 2016: 116 · 2015: 24
FWCI: 230.94
Type: journal article
Language: en
Open access: closed
Licence: ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html (vor)
Locations: IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
PubMed (publishedVersion)
The Royal Society of Chemistry’s Journals, Books and Databases (The Royal Society of Chemistry) (submittedVersion)
Funders: NSFC (61005027, 61370036)
ICT R&D programs of MSIP/IITP (10047078, 14-824-09-006)
National Science Foundation (1149783)
IIS (1152576)
Topics: Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Video Surveillance and Tracking Methods
Physical Sciences › Engineering › Automotive Engineering › Autonomous Vehicle Technology and Safety
Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords: object tracking, benchmark dataset, tracking algorithm evaluation, online tracking, performance evaluation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Object Tracking Benchmark

Abstract

Overview

AI snapshot

TL;DR

Key findings

Methods

Limitations

Integrity signals

Verification

Claims the paper makes

Claims other papers make about this one

Citations

Cites

Cited by

Related

Metadata

Record

About these checks