Learning Support Correlation Filters for Visual Tracking

Zuo, Wangmeng; Wu, Xiaohe; Lin, Liang; Zhang, Lei; Yang, Ming–Hsuan

doi:10.1109/tpami.2018.2829180

Cited by 102 publications

(80 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…For SVOS methods, the target object(s) is provided in the first frame and tracked automatically [60,8,5,68,2,69,64,71] or interactively by users [1] in the subsequent frames. Numerous algorithms were proposed based on graphical models [54], object proposals [46], supertrajectories [61], etc.…”

Section: Video Object Segmentationmentioning

confidence: 99%

See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks

Yang

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

469

226

View full text Add to dashboard Cite

We introduce a novel network, called CO-attention Siamese Network (COSNet), to address the unsupervised video object segmentation task from a holistic view. We emphasize the importance of inherent correlation among video frames and incorporate a global co-attention mechanism to improve further the state-of-the-art deep learning based solutions that primarily focus on learning discriminative foreground representations over appearance and motion in short-term temporal segments. The co-attention layers in our network provide efficient and competent stages for capturing global correlations and scene context by jointly computing and appending co-attention responses into a joint feature space. We train COSNet with pairs of video frames, which naturally augments training data and allows increased learning capacity. During the segmentation stage, the co-attention model encodes useful information by processing multiple reference frames together, which is leveraged to infer the frequently reappearing and salient foreground objects better. We propose a unified and end-to-end trainable framework where different co-attention variants can be derived for mining the rich context within videos. Our extensive experiments over three large benchmarks manifest that COSNet outperforms the current alternatives by a large margin.

show abstract

Section: Video Object Segmentationmentioning

confidence: 99%

See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks

Yang

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

469

226

View full text Add to dashboard Cite

show abstract

“…The Fast Fourier Transform (FFT) employed for mapping variables across the spatial and spatial frequency domains enables further acceleration of the computation. The computational efficacy of DCF received a wide attention, stimulating improved formulations in multi-response fusion (Staple) [27], circulant sparse representation (CST) [28], support vector filters (SCF) [29], and so on.…”

Section: Dcf Tracking Formulationsmentioning

confidence: 99%

An accelerated correlation filter tracker

Xu¹,

Feng²,

Wu³

et al. 2020

Pattern Recognition

View full text Add to dashboard Cite

Recent visual object tracking methods have witnessed a continuous improvement in the state-of-the-art with the development of efficient discriminative correlation filters (DCF) and robust deep neural network features. Despite the outstanding performance achieved by the above combination, existing advanced trackers suffer from the burden of high computational complexity of the deep feature extraction and online model learning. We propose an accelerated ADMM optimisation method obtained by adding a momentum to the optimisation sequence iterates, and by relaxing the impact of the error between DCF parameters and their norm. The proposed optimisation method is applied to an innovative formulation of the DCF design, which seeks the most discriminative spatially regularised feature channels. A further speed up is achieved by an adaptive initialisation of the filter optimisation process. The significantly increased convergence of the DCF filter is demonstrated by establishing the optimisation process equivalence with a continuous dynamical system for which the convergence properties can readily be derived. The experimental results obtained on several well-known benchmarking datasets demonstrate the efficiency and robustness of the proposed ACFT method, with a tracking accuracy comparable to the start-of-the-art trackers.

show abstract

“…The position coordinates of this region in each frame is recorded, and the variation of the region is

d_{rough}^{((), t)}

. SCF is a fast and robust tracking algorithmand can track an object region well, so SCF is employed to obtain

d_{rough}^{((), t)}

in this study. To accomplish SCF, an n × n square patch (original) containing the object of interest is first selected.…”

Section: Identification Algorithmmentioning

confidence: 99%

“…These methods are suitable for tracking a target area. The KCF (kernelized correlation filters) and SCF methods are effective CF algorithm.…”

Section: Introductionmentioning

confidence: 99%

“…The KCF (kernelized correlation filters) algorithm has been used widely in the video tracking field in recent years due to its high speed and accuracy. SCF (used in this work), proposed by Zuo et al, is an improved version of KCF. One of the differences between the algorithms is that the kernel ridge regression optimization problem in KCF (kernelized correlation filters) is changed to a classification problem of SVM in SCF.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Video-based multiscale identification approach for tower vibration of a cable-stayed bridge model under earthquake ground motions

Zhao

Bao

Guan

et al. 2019

Struct Control Health Monit

Self Cite

View full text Add to dashboard Cite

Summary A camera can theoretically capture visual information of all target surfaces within its field of view. This capability allows a camera to act as a sensor for determining motion or deformation of targets at appropriate distances. Considering that there are so many cameras in public society (including cameras in smartphones) that can record the vibration of structures in an earthquake, this capability will be very helpful for assessing structural seismic damage post‐earthquake. To this end, this paper proposes a two‐step combination of the SCF (support correlation filters) algorithm with the KLT (Kanade‐Lucas‐Tomasi) algorithm for displacement identification of cable‐stayed bridges from video with higher accuracy over SCF only and higher robustness over KLT only. The SCF algorithm can robustly and quickly track the target in the video, and the KLT algorithm can accurately track the pixel locations of targets. The shaking table test of a scale cable‐stayed bridge model was conducted, and the vibration of the tower was recorded by a camera. The linear and nonlinear displacement responses of the tower under different earthquake ground motions were identified. The accuracy of the identified displacement response was validated through comparison with measurements from laser displacement transducers.

show abstract

Learning Support Correlation Filters for Visual Tracking

Cited by 102 publications

References 64 publications

See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks

See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks

An accelerated correlation filter tracker

Video-based multiscale identification approach for tower vibration of a cable-stayed bridge model under earthquake ground motions

Contact Info

Product

Resources

About