Back

Object Tracking Benchmark

Yi WuJongwoo LimMing-Hsuan Yang

article2015TPAMI3,221 citations

Establishes a standardized visual tracking evaluation platform featuring 50 fully annotated video sequences categorized by 11 challenge attributes, an integrated code library of 29 algorithms, and systematic spatial and temporal perturbation metrics to assess tracking performance.

Listen

Object tracking remains a core challenge in computer vision for applications such as surveillance and medical imaging, yet inconsistent datasets, varying initial conditions, and non-uniform code have made it difficult to compare algorithms reliably or identify what drives robust performance.

This paper set out to create a standardized benchmark that would allow large-scale, fair evaluation of recent online single-target trackers and reveal which design choices matter most under realistic conditions.

The authors assembled a library of 29 publicly available trackers with uniform input and output formats, collected and fully annotated 50 video sequences with ground-truth bounding boxes plus 11 attributes that commonly affect tracking, and ran extensive tests using precision plots at a 20-pixel threshold and success plots measured by area under the curve. They evaluated each tracker more than 660,000 times by applying one-pass evaluation plus two new robustness protocols that perturb the starting frame and the initial bounding-box location or scale.

The evaluation shows that SCM, Struck, and ASLA rank at the top overall, with SCM leading in one-pass tests and Struck proving more stable when initialization varies. Local sparse representations outperform holistic sparse templates on occlusion and deformation; trackers that explicitly exploit background context or use structured output learning handle partial occlusions better; and dense-sampling methods with discriminative models cope more effectively with fast motion than particle-filter approaches. Performance drops noticeably when the initial box is enlarged by 20 percent or shifted, confirming that most trackers remain sensitive to scale and position errors introduced by detectors. Finally, the results indicate that motion or dynamic models receive far less attention than representation and search mechanisms, even though they strongly influence both accuracy and efficiency on fast or abrupt motion.

These findings imply that future trackers can gain the most by combining local appearance models, implicit or explicit background modeling, and stronger motion prediction, rather than refining any single component in isolation. The benchmark itself supplies a reusable platform that removes much of the ambiguity that has hindered progress.

The authors recommend extending both the dataset and the code library with additional sequences and trackers, and they note that improving dynamic models and testing under detector-driven initialization would yield the next practical advances. The work is limited to online single-target tracking with publicly released code and to the 50 sequences chosen; results could shift with different attribute distributions or parameter settings outside the defaults supplied by each method. The scale of the experiments and the consistency of the evaluation protocols give high confidence in the relative rankings and component analysis within the stated scope.

Hanyang UniversityUniversity of California, Mercedobject trackingObject Tracking Benchmarkvisual trackingspatial robustness evaluation
Wu et al (2015).pdf
  • Paper: The Pascal Visual Object Classes Challenge: A Retrospective, M. Everingham et al. (2014). Reading the PASCAL VOC retrospective first provides the foundational benchmark context and evaluation standards necessary for understanding large-scale object detection challenges.
  • Paper: ImageNet Large Scale Visual Recognition Challenge, Olga Russakovsky et al. (2014). Understanding the ImageNet Large Scale Visual Recognition Challenge offers crucial background on the evolution of standardized visual benchmarking that preceded contemporary tracking and detection libraries.
Cover for Object Tracking Benchmark

Abstract

Object tracking is one of the most important components in numerous applications of computer vision. While much progress has been made in recent years with efforts on sharing code and datasets, it is of great importance to develop a library and benchmark to gauge the state of the art. After briefly reviewing recent advances of online object tracking, we carry out large scale experiments with various evaluation criteria to understand how these algorithms perform. The test image sequences are annotated with different attributes for performance evaluation and analysis. By analyzing quantitative results, we identify effective approaches for robust tracking and provide potential future research directions in this field.

Topics

object tracking

Keywords

Object Tracking Benchmarkvisual trackingspatial robustness evaluationtemporal robustnessappearance modelscovariance trackingL1 trackerocclusion detectionoptical flow

Table of Contents

  • 1. Introduction
  • 2. Related Work
  • 3. Evaluated Algorithms and Datasets
  • 4. Evaluation Methodology
  • 5. Evaluation Results
  • 5.1. Overall Performance
  • 5.2. Attribute-based Performance Analysis
  • 5.3. Initialization with Different Scale
  • 6. Concluding Remarks
  • References

Knowls

  1. Knowl 1 — Precision and Success Rate Evaluation Metrics in Visual Tracking

    model/method

    Visual tracking performance is quantitatively evaluated using two primary curves across continuous error thresholds:

    1. Precision Plot (Center Location Error): Evaluates tracking accuracy based on the Euclidean distance between the tracked target center location and the manually labeled ground-truth center. For a video sequence, the precision plot shows the percentage of frames whose estimated center is within a distance threshold from the ground truth. The standardized representative ranking metric is the precision score at a threshold of 2020 pixels.

    2. Success Plot (Bounding Box Overlap): Evaluates spatial alignment based on the intersection-over-union (IoU) overlap score SS between the predicted bounding box rtr_t and the ground truth bounding box rar_a:

    S=∣rt∩ra∣∣rt∪ra∣S = \frac{|r_t \cap r_a|}{|r_t \cup r_a|}

    where ∩\cap and ∪\cup denote the spatial intersection and union of bounding box regions, respectively, and ∣⋅∣|\cdot| represents the number of pixels within that region. Given an overlap threshold to∈[0,1]t_o \in [0, 1], a frame is counted as successful if S>toS > t_o. The success plot illustrates the success rate across all to∈[0,1]t_o \in [0, 1]. Trackers are ranked based on the Area Under the Curve (AUC) of the success plot rather than a single cutoff score (such as to=0.5t_o = 0.5).

  2. Knowl 2 — Robustness Evaluation Protocols: OPE, SRE, and TRE

    model/method

    To evaluate tracker sensitivity to initialization conditions, three distinct evaluation protocols are defined:

    1. One-Pass Evaluation (OPE): The standard tracking protocol where a tracker is initialized with the ground-truth target bounding box in the first frame and runs forward through to the final frame without re-initialization.

    2. Spatial Robustness Evaluation (SRE): Evaluates sensitivity to spatial bounding box errors (simulating detector noise). The initial bounding box in the first frame is perturbed 1212 times:

      • 88 spatial shifts (4 center shifts and 4 corner shifts) shifted by 10%10\% of the target's width and height.
      • 44 scale variations scaling the initial bounding box dimensions by factors of 0.80.8, 0.90.9, 1.11.1, and 1.21.2 relative to the ground truth. Performance is averaged across all 1212 trials per sequence.

    3. Temporal Robustness Evaluation (TRE): Evaluates tracker robustness across different starting points. Each video sequence is partitioned into 2020 uniformly spaced start frames. In each test, the tracker is initialized from the ground truth bounding box at that frame and executed until the end of the sequence (generating 2020 distinct sequence segments per video). Performance statistics are averaged across all segment evaluations.

  3. Knowl 3 — Visual Tracking Sequence Attribute Taxonomy

    definition

    Tracking test sequences are categorized using 1111 visual attributes to evaluate tracker performance across distinct tracking challenges:

    • Illumination Variation (IV): The illumination of the target region changes significantly.
    • Scale Variation (SV): The ratio between the bounding box area in the first frame and the current frame falls outside [1/ts,ts][1/t_s, t_s], with threshold ts=2t_s = 2.
    • Occlusion (OCC): The target is partially or fully occluded by background elements or other objects.
    • Deformation (DEF): Non-rigid deformation of the target object.
    • Motion Blur (MB): The target region is blurred due to fast target or camera motion.
    • Fast Motion (FM): Ground-truth target motion between consecutive frames exceeds tm=20t_m = 20 pixels.
    • In-Plane Rotation (IPR): The target rotates in the 2D image plane.
    • Out-of-Plane Rotation (OPR): The target rotates in 3D out of the image plane.
    • Out-of-View (OV): Some portion of the target region leaves the video frame view.
    • Background Clutters (BC): Background regions adjacent to the target have similar color or texture characteristics as the target.
    • Low Resolution (LR): Target ground-truth bounding box area contains fewer than tr=400t_r = 400 pixels.

  4. Knowl 4 — Evaluated Tracking Algorithms and System Specifications

    data/table

    The benchmark evaluates 29 tracking algorithms categorized by their target representation scheme (Local: L, Holistic: H; Template: T, Intensity Histogram: IH, Binary Pattern: BP, PCA, Sparse PCA: SPCA, Sparse Representation: SR; Discriminative Model: DM, Generative Model: GM), search mechanism (Particle Filter: PF, Markov Chain Monte Carlo: MCMC, Local Optimum Search: LOS, Dense Sampling: DS), model update (MU: Yes/No), programming language, and tracking speed in Frames Per Second (FPS) on an Intel Core i7 3770 CPU (3.4 GHz):

    Method Representation Search MU Code FPS
    CPF L, IH PF N C 109
    LOT L, color PF Y M 0.70
    IVT H, PCA, GM PF Y MC 33.4
    ASLA L, SR, GM PF Y MC 8.5
    SCM L, SR, GM+DM PF Y MC 0.51
    L1APG H, SR, GM PF Y MC 2.0
    MTT H, SR, GM PF Y M 1.0
    VTD H, SPCA, GM MCMC Y MC-E 5.7
    VTS L, SPCA, GM MCMC Y MC-E 5.7
    LSK L, SR, GM LOS Y M-E 5.5
    ORIA H, T, GM LOS Y M 9.0
    DFT L, T LOS Y M 13.2
    KMS H, IH LOS N C 3159
    SMS H, IH LOS N C 19.2
    VR-V H, color LOS Y MC 109
    Frag L, IH DS N C 6.3
    OAB H, Haar, DM DS Y C 22.4
    SemiT H, Haar, DM DS Y C 11.2
    BSBT H, Haar, DM DS Y C 7.0
    MIL H, Haar, DM DS Y C 38.1
    CT H, Haar, DM DS Y MC 64.4
    TLD L, BP, DM DS Y MC 28.1
    Struck H, Haar, DM DS Y C 20.2
    CSK H, T, DM DS Y M 362
    CXT H, BP, DM DS Y C 15.3

    Code languages indicate C/C++ (C), Matlab (M), mixed C/C++ and Matlab (MC), or executable binaries (-E). Algorithms span a spectrum of trade-offs between representation complexity (e.g., SCM at 0.51 FPS vs. CSK at 362 FPS) and classification framework.

  5. Knowl 5 — Overall Tracking Performance Discrepancies Across Evaluation Protocols

    empirical result

    Tracking rankings and success rates differ substantially across evaluation protocols:

    • Protocol Discrepancies: In OPE, SCM achieves the highest AUC score (0.4990.499), outperforming Struck (0.4730.473) by 2.6%2.6\%. However, in SRE, Struck ranks highest (0.4350.435) while SCM drops to second (0.4210.421, a 1.9%1.9\% difference), showing that single-pass evaluation (OPE) does not reliably reflect initialization robustness.
    • Protocol Score Trends: Average TRE success scores are higher than OPE across all trackers because segment length decreases over later segments, and trackers generally perform better on shorter sequences. Conversely, average SRE success scores are lower than OPE because spatial initialization errors introduce background pixels into initial appearance models, leading to accumulated drift.
    • Scale Limitation of Struck: In SRE and TRE success plots, Struck exhibits higher success rates than SCM and ASLA at small overlap thresholds tot_o, but lower success rates at high overlap thresholds (to>0.6t_o > 0.6). This occurs because Struck outputs bounding boxes with fixed initial scale and only estimates 2D translational position, penalizing IoU at tight thresholds.

  6. Knowl 6 — Tracker Performance on Fast Motion, Occlusion, and Scale Variation Subsets

    empirical result

    Performance across attribute-specific subsets reveals distinct architectural strengths and bottlenecks under Spatial Robustness Evaluation (SRE):

    • Fast Motion (FM): Dense-sampling discriminative trackers (Struck, TLD, CXT) significantly outperform stochastic particle filter trackers. Large search windows coupled with discriminative classifiers allow them to capture abrupt displacements. Stochastic particle filter trackers with high overall rankings (such as SCM and ASLA) degrade due to narrow Gaussian dynamic motion models that require prohibitively large particle counts to track rapid jumps.
    • Occlusion (OCC): Trackers utilizing structured learning or local sparse representations (Struck, SCM, TLD, LSK, ASLA) achieve the highest success and precision scores. Local representation allows uncorrupted target regions to maintain track despite partial occlusion.
    • Scale Variation (SV): Trackers incorporating affine motion models within particle filtering (such as ASLA and SCM) outperform translational-only trackers by adapting the bounding box geometry to scale changes.

  7. Knowl 7 — Tracker Sensitivity to Initial Bounding Box Scale Variations

    empirical result

    Evaluating tracker performance under initial bounding box scale perturbation factors ranging from 0.8×0.8\times to 1.2×1.2\times demonstrates specific representation sensitivities:

    • Sensitivity to Scale Expansion (1.2×1.2\times): Overall tracking performance drops sharply as scale factor increases to 1.2×1.2\times because excess background pixels contaminate the initial appearance model. TLD, CXT, DFT, and LOT exhibit the steepest performance declines, indicating high susceptibility to background noise.
    • Sensitivity to Scale Contraction (0.8×0.8\times): Certain generative sparse trackers (such as L1APG and MTT) achieve higher tracking accuracy with smaller initial scale (0.8×0.8\times or 0.9×0.9\times). This occurs because images are downsampled to a small canonical template size, allowing smaller bounding boxes to preserve more high-frequency target detail upon normalization.
    • Robustness of Haar-Like Features: Trackers using Haar-like features (Struck, OAB, SemiT, BSBT) remain stable or slightly improve when the bounding box is enlarged, as rectangular summation pooling attenuates uncorrelated background clutter. Struck shows the lowest overall sensitivity to initialization scale perturbations.

  8. Knowl 8 — Local vs. Holistic Sparse Representations in Online Tracking

    empirical result

    Comparison of sparse-representation-based trackers shows a distinct performance division between local patch-based models and holistic template models:

    • Performance Gap: Local sparse appearance models (SCM, ASLA, LSK) consistently outperform holistic sparse template models (MTT, L1APG) across OPE, SRE, and TRE protocols.
    • Mechanism: Holistic sparse templates represent the entire bounding box via global vectors and are easily corrupted when appearance changes non-uniformly across the target. Local sparse models segment the object into local patches and reconstruct them independently, making them robust to local deformations and partial occlusions.
    • Alignment-Pooling in ASLA: ASLA exhibits the lowest AUC degradation between OPE and SRE among the top 5 trackers, demonstrating that alignment-pooling over local sparse representations provides robustness against spatial misalignment and background clutter.

  9. Knowl 9 — Speed and Computational Complexity Across Tracking Paradigms

    empirical result

    Benchmarking inference speed across tracking paradigms reveals a wide computational spectrum:

    • Fourier/Circulant Structure: CSK achieves the highest real-time speed among high-performing trackers (362362 FPS) by exploiting circulant matrix structures in the frequency domain via Fast Fourier Transforms (FFT).
    • Mean-Shift and Local Search: Histogram-based local search trackers achieve the highest overall frame rates (KMS at 3,1593,159 FPS, CPF and VR-V at 109109 FPS), but lack adaptive update complexity.
    • Sparse Representation Bottlenecks: Trackers relying on online particle-filter-based ℓ1\ell_1-minimization or local patch dictionary solving represent the slowest class (SCM at 0.510.51 FPS, LOT at 0.700.70 FPS, MTT at 1.01.0 FPS, L1APG at 2.02.0 FPS), creating a significant efficiency bottleneck despite their tracking accuracy.

Coverage note — No substantial contributed material was omitted. Specific algorithm implementation details of external trackers (e.g., individual equations for IVT, MIL, Struck) were excluded as they constitute referenced prior works rather than contributions of this benchmark.

References

  1. 1.A. Adam, E. Rivlin, and I. Shimshoni. Robust Fragments-based Tracking using the Integral Histogram. In CVPR, 2006.
  2. 2.N. Alt, S. Hinterstoisser, and N. Navab. Rapid Selection of Reliable Templates for Visual Tracking. In CVPR, 2010.
  3. 3.S. Avidan. Support Vector Tracking. PAMI, 26(8):1064–1072, 2004.
  4. 4.S. Avidan. Ensemble Tracking. PAMI, 29(2):261–271, 2008.
  5. 5.B. Babenko, M.-H. Yang, and S. Belongie. Visual Tracking with Online Multiple Instance Learning. In CVPR, 2009.
  6. 6.B. Babenko, M.-H. Yang, and S. Belongie. Robust Object Tracking with Online Multiple Instance Learning. PAMI, 33(7):1619–1632, 2011.
  7. 7.Y. Bai and M. Tang. Robust Tracking via Weakly Supervised Ranking SVM. In CVPR, 2012.
  8. 8.S. Baker and I. Matthews. Lucas-Kanade 20 Years On: A Unifying Framework. IJCV, 56(3):221–255, 2004.
  9. 9.S. Baker, S. Roth, D. Scharstein, M. J. Black, J. Lewis, and R. Szeliski. A Database and Evaluation Methodology for Optical Flow. In ICCV, 2007.
  10. 10.C. Bao, Y. Wu, H. Ling, and H. Ji. Real Time Robust L1 Tracker Using Accelerated Proximal Gradient Approach. In CVPR, 2012.
  11. 11.M. J. Black. EigenTracking: Robust Matching and Tracking of Articulated Objects Using a View-Based Representation. IJCV, 26(1):63–84, 1998.
  12. 12.K. Cannons. A Review of Visual Tracking. Technical Report CSE-2008-07, York University, Canada, 2008.
  13. 13.R. Collins. Mean-shift Blob Tracking through Scale Space. In CVPR, 2003.
  14. 14.R. Collins, X. Zhou, and S. K. Teh. An Open Source Tracking Testbed and Evaluation Web Site. In PETS, 2005.
  15. 15.R. T. Collins, Y. Liu, and M. Leordeanu. Online Selection of Discriminative Tracking Features. PAMI, 27(10):1631–1643, 2005.
  16. 16.D. Comaniciu, V. Ramesh, and P. Meer. Kernel-Based Object Tracking. PAMI, 25(5):564–577, 2003.
  17. 17.N. Dalal and B. Triggs. Histograms of Oriented Gradients for Human Detection. In CVPR, 2005.
  18. 18.T. B. Dinh, N. Vo, and G. Medioni. Context Tracker: Exploring Supporters and Distracters in Unconstrained Environments. In CVPR, 2011.
  19. 19.J. Fan, X. Shen, and Y. Wu. Scribble Tracker: A Matting-based Approach for Robust Tracking. PAMI, 34(8):1633–1644, 2012.
  20. 20.J. Fan, Y. Wu, and S. Dai. Discriminative Spatial Attention for Robust Tracking. In ECCV, 2010.
  21. 21.R. B. Fisher. The PETS04 Surveillance Ground-Truth Data Sets. In PETS, 2004.
  22. 22.H. Grabner, M. Grabner, and H. Bischof. Real-Time Tracking via On-line Boosting. In BMVC, 2006.
  23. 23.H. Grabner, C. Leistner, and H. Bischof. Semi-supervised On-Line Boosting for Robust Tracking. In ECCV, 2008.
  24. 24.H. Grabner, J. Matas, L. V. Gool, and P. Cattin. Tracking the Invisible: Learning Where the Object Might be. In CVPR, 2010.
  25. 25.G. D. Hager and P. N. Belhumeur. Efficient Region Tracking With Parametric Models of Geometry and Illumination. PAMI, 20(10):1025–1039, 1998.
  26. 26.S. Hare, A. Saffari, and P. H. S. Torr. Struck: Structured Output Tracking with Kernels. In ICCV, 2011.
  27. 27.F. Henriques, R. Caseiro, P. Martins, and J. Batista. Exploiting the Circulant Structure of Tracking-by-Detection with Kernels. In ECCV, 2012.
  28. 28.M. Isard and A. Blake. CONDENSATION–Conditional Density Propagation for Visual Tracking. IJCV, 29(1):5–28, 1998.
  29. 29.A. D. Jepson, D. J. Fleet, and T. F. El-Maraghi. Robust Online Appearance Models for Visual Tracking. PAMI, 25(10):1296–1311, 2003.
  30. 30.X. Jia, H. Lu, and M.-H. Yang. Visual Tracking via Adaptive Structural Local Sparse Appearance Model. In CVPR, 2012.
  31. 31.Z. Kalal, J. Matas, and K. Mikolajczyk. P-N Learning: Bootstrapping Binary Classifiers by Structural Constraints. In CVPR, 2010.
  32. 32.J. Kwon and K. M. Lee. Tracking of a Non-Rigid Object via Patch-based Dynamic Appearance Modeling and Adaptive Basin Hopping Monte Carlo Sampling. In CVPR, 2009.
  33. 33.J. Kwon and K. M. Lee. Visual Tracking Decomposition. In CVPR, 2010.
  34. 34.J. Kwon and K. M. Lee. Tracking by Sampling Trackers. In ICCV, 2011.
  35. 35.X. Li, W. Hu, C. Shen, Z. Zhang, A. Dick, and A. Hengel. A Survey of Appearance Models in Visual Object Tracking. TIST, 2013, in press.
  36. 36.B. Liu, J. Huang, L. Yang, and C. Kulikowsk. Robust Tracking using Local Sparse Appearance Model and K-Selection. In CVPR, 2011.
  37. 37.B. D. Lucas and T. Kanade. An Iterative Image Registration Technique with An Application to Stereo Vision. In IJCAI, 1981.
  38. 38.D. R. Martin, C. C. Fowlkes, and J. Malik. Learning to Detect Natural Image Boundaries Using Local Brightness, Color, and Texture Cues. PAMI, 26(5):530–49, 2004.
  39. 39.I. Matthews, T. Ishikawa, and S. Baker. The Template Update Problem. PAMI, 26(6):810–815, 2004.
  40. 40.X. Mei and H. Ling. Robust Visual Tracking using L1 Minimization. In ICCV, 2009.
  41. 41.X. Mei, H. Ling, Y. Wu, E. Blasch, and L. Bai. Minimum Error Bounded Efficient L1 Tracker with Occlusion Detection. In CVPR, 2011.
  42. 42.X. Mei, H. Ling, Y. Wu, E. Blasch, and L. Bai. Efficient Minimum Error Bounded Particle Resampling L1 Tracker with Occlusion Detection. TIP, 2013, in press.
  43. 43.S. Oron, A. Bar-Hillel, D. Levi, and S. Avidan. Locally Orderless Tracking. In CVPR, 2012.
  44. 44.P. Pérez, C. Hue, J. Vermaak, and M. Gangnet. Color-Based Probabilistic Tracking. In ECCV, 2002.
  45. 45.P. Phillips, H. Moon, S. Rizvi, and P. Rauss. The FERET Evaluation Methodology for Face-Recognition Algorithms. PAMI, 22(10):1090–1104, 2000.
  46. 46.F. Porikli, O. Tuzel, and P. Meer. Covariance Tracking using Model Update based on Lie Algebra. In CVPR, 2006.
  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.
  48. 48.J. Santner, C. Leistner, A. Saffari, T. Pock, and H. Bischof. PROST: Parallel Robust Online Simple Tracking. In CVPR, 2010.
  49. 49.L. Sevilla-Lara and E. Learned-Miller. Distribution Fields for Tracking. In CVPR, 2012.
  50. 50.S. Stalder, H. Grabner, and L. van Gool. Beyond Semi-Supervised Tracking: Tracking Should Be as Simple as Detection, but not Simpler than Recognition. In ICCV Workshop, 2009.
  51. 51.B. Stenger, T. Woodley, and R. Cipolla. Learning to Track with Multiple Observers. In CVPR, 2009.
  52. 52.F. Tang, S. Brennan, Q. Zhao, and H. Tao. Co-Tracking Using Semi-Supervised Support Vector Machines. CVPR, 2007.
  53. 53.O. Tuzel, F. Porikli, and P. Meer. Region Covariance: A Fast Descriptor for Detection and Classification. In ECCV, 2006.
  54. 54.P. Viola and M. J. Jones. Robust Real-Time Face Detection. IJCV, 57(2):137–154, 2004.
  55. 55.D. Wang, H. Lu, and M.-H. Yang. Online Object Tracking with Sparse Prototypes. TIP, 22(1):314–325, 2013.
  56. 56.Y. Wu, J. Cheng, J. Wang, H. Lu, J. Wang, H. Ling, E. Blasch, and L. Bai. Real-time Probabilistic Covariance Tracking with Efficient Model Update. TIP, 21(5):2824–2837, 2012.
  57. 57.Y. Wu, H. Ling, J. Yu, F. Li, X. Mei, and E. Cheng. Blurred Target Tracking by Blur-driven Tracker. In ICCV, 2011.
  58. 58.Y. Wu, B. Shen, and H. Ling. Online Robust Image Alignment via Iterative Convex Optimization. In CVPR, 2012.
  59. 59.M. Yang, Y. Wu, and G. Hua. Context-Aware Visual Tracking. PAMI, 31(7):1195–1209, 2008.
  60. 60.A. Yilmaz, O. Javed, and M. Shah. Object Tracking: A Survey. ACM Computing Surveys, 38(4):1–45, 2006.
  61. 61.J. H. Yoon, D. Y. Kim, and K.-J. Yoon. Visual Tracking via Adaptive Tracker Selection with Multiple Features. In ECCV, 2012.
  62. 62.L. Yuan, A. Haizhou, T. Yamashita, L. Shihong, and M. Kawade. Tracking in Low Frame Rate Video: A Cascade Particle Filter with Discriminative Observers of Different Life Spans. PAMI, 30(10):1728–1740, 2008.
  63. 63.K. Zhang, L. Zhang, and M.-H. Yang. Real-time Compressive Tracking. In ECCV, 2012.
  64. 64.T. Zhang, B. Ghanem, S. Liu, and N. Ahuja. Robust Visual Tracking via Multi-task Sparse Learning. In CVPR, 2012.
  65. 65.W. Zhong, H. Lu, and M.-H. Yang. Robust Object Tracking via Sparsity-based Collaborative Model. In CVPR, 2012.

Citation

MLA
Wu, Y., et al. “Object Tracking Benchmark”. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 37, no. 9, 2015, pp. 1834–48, https://doi.org/10.1109/TPAMI.2014.2388226.
APA
Wu, Y., Lim, J., & Yang, M.-H. (2015). Object Tracking Benchmark. IEEE Transactions on Pattern Analysis and Machine Intelligence, 37(9), 1834–1848. https://doi.org/10.1109/TPAMI.2014.2388226
Chicago
Wu, Y., J. Lim, and M.-H. Yang. 2015. “Object Tracking Benchmark”. IEEE Transactions on Pattern Analysis and Machine Intelligence 37 (9): 1834–48. https://doi.org/10.1109/TPAMI.2014.2388226.
Harvard
Wu, Y., Lim, J. and Yang, M.-H. (2015) “Object Tracking Benchmark”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 37(9), pp. 1834–1848. Available at: https://doi.org/10.1109/TPAMI.2014.2388226.
Vancouver
1. Wu Y, Lim J, Yang M-H (2015) Object Tracking Benchmark. IEEE Transactions on Pattern Analysis and Machine Intelligence 37:1834–1848

BibTeX

@article{Wu_2015, title={Object Tracking Benchmark}, volume={37}, ISSN={2160-9292}, url={http://dx.doi.org/10.1109/TPAMI.2014.2388226}, DOI={10.1109/tpami.2014.2388226}, number={9}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, publisher={Institute of Electrical and Electronics Engineers (IEEE)}, author={Wu, Yi and Lim, Jongwoo and Yang, Ming-Hsuan}, year={2015}, month=Sept, pages={1834–1848} }
Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF