Masked-RPCA: Moving Object Detection With an Overlaying Model

Khalilian-Gourtani, Amirhossein; Minaee, Shervin; Wang, Yao

doi:10.1109/ojsp.2020.3039325

Cited by 6 publications

(12 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…well fits the condition of the RPCA method [28], where represents the frequency domain of rotating part which is also a low-rank matrix, represents the frequency domain of static part including the target and the jamming signal which is also a sparse matrix, and represents a noise matrix.…”

Section: Analysis Of the Fd-rpca-bss Methodsmentioning

confidence: 69%

Mainlobe jamming suppression via improved BSS method for rotated array radar

Zhang

Biao

et al. 2022

J. of Syst. Eng. Electron.

View full text Add to dashboard Cite

This study deals with the problem of mainlobe jamming suppression for rotated array radar. The interference becomes spatially nonstationary while the radar array rotates, which causes the mismatch between the weight and the snapshots and thus the loss of target signal to noise ratio (SNR) of pulse compression. In this paper, we explore the spatial divergence of interference sources and consider the rotated array radar anti-mainlobe jamming problem as a generalized rotated array mixed signal (RAMS) model firstly. Then the corresponding algorithm improved blind source separation (BSS) using the frequency domain of robust principal component analysis (FD-RPCA-BSS) is proposed based on the established rotating model. It can eliminate the influence of the rotating parts and address the problem of loss of SNR . Finally, the measured peakto-average power ratio (PAPR) of each separated channel is performed to identify the target echo channel among the separated channels. Simulation results show that the proposed method is practically feasible and can suppress the mainlobe jamming with lower loss of SNR.

show abstract

Section: Analysis Of the Fd-rpca-bss Methodsmentioning

confidence: 69%

Mainlobe jamming suppression via improved BSS method for rotated array radar

Zhang

Biao

et al. 2022

J. of Syst. Eng. Electron.

View full text Add to dashboard Cite

show abstract

“…The Masked-RPCA (MRPCA) method [15], [47] changes the problem of foreground separation to a foreground detection problem, where a sparse foreground mask M ∈ {0, 1} hw×t is predicted instead of the typical foreground S of RPCA. In fact, using a simple threshold on S to identify foreground pixels may be insufficient when the pixel difference with the background is low, or in video with high disturbances.…”

Section: Masked Rpcamentioning

confidence: 99%

“…The sparsity of M and the correlation in time with M is formulated as a 1 -1 -minimization penalty term, λ 1 , λ 2 are regularization parameters and • denotes the Hadamard product. Problem (14) differs from the Masked-RPCA model [47] since our model uses a side-information branch and measurement operators H 1 and H 2 , which create learnable weights in the deep unfolding steps [36], [37]. We then follow a similar approach to [15] by reformulating the non-convex problem (14) in the augmented Lagrangian form with a dual variable U.…”

Section: Masked Rpcamentioning

confidence: 99%

Interpretable Neural Networks for Video Separation: Deep Unfolding RPCA with Foreground Masking

Joukovsky¹,

Deligiannis²,

Eldar³

2023

Preprint

View full text Add to dashboard Cite

This paper presents two deep unfolding neural networks for the simultaneous tasks of background subtraction and foreground detection in video. Unlike conventional neural networks based on deep feature extraction, we incorporate domain-knowledge models by considering a masked variation of the robust principal component analysis problem (RPCA). With this approach, we separate video clips into low-rank and sparse components, respectively corresponding to the backgrounds and foreground masks indicating the presence of moving objects. Our models, coined ROMAN-S and ROMAN-R, map the iterations of two alternating direction of multipliers methods (ADMM) to trainable convolutional layers, and the proximal operators are mapped to non-linear activation functions with trainable thresholds. This approach leads to lightweight networks with enhanced interpretability that can be trained on few data. In ROMAN-S, the correlation in time of successive binary masks is controlled with a side-information scheme based on L1-L1 minimization. ROMAN-R enhances the foreground detection by learning a dictionary of atoms to represent the moving foreground in a high-dimensional feature space and by using reweighted-L1-L1 minimization. Experiments are conducted on both synthetic and real video datasets and comparisons are made with existing deep unfolding RPCA neural networks, which do not use a mask formulation for the foreground. The models are also compared to a U-Net baseline. Results show that our proposed models outperform other deep unfolding models, as well as the untrained optimization algorithms. ROMAN-R, in particular, is competitive with the U-Net baseline for foreground detection, with the additional advantage of providing video backgrounds and requiring substantially fewer training parameters and smaller training sets.

show abstract

“…The small singular value is compressed with a large amplitude, which further reduces the dynamic data component such as slow moving target in the background. Based on the above two discussions, Frobenius norm and weighted kernel norm are introduced into the model of formula (1). The improved model is as follows:…”

Section: Problem Formulationmentioning

confidence: 99%

“…Moving object detection [1,2] is a key step in many computer vision applications. The results of moving object detection are often affected by factors such as slow foreground objects, rain and snow weather, and dynamic backgrounds.…”

Section: Introductionmentioning

confidence: 99%

Moving Object Detection Based on Weighted Kernel Norm and Saliency Constrained RPCA

Zhao

Zheng

Liu

et al. 2022

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

Focusing on the issue that the accuracy of object detection is reduced when the foreground target moves slowly and there is interference in the background, a moving object detection method based on weighted kernel norm and saliency constraint RPCA is proposed. In the new method, the weighted kernel norm is used to restore a relatively clean background, which is helpful to separate the slow moving target from the background. The l 1 norm is used to constrain the sparsity of moving objects, and the Frobenius norm is used to detect noise, and a saliency constraint is introduced to detect slow-moving objects. Experiments show that this method can effectively deal with the problems of slow foreground motion and background interference. Compared with the suboptimal algorithm, the average measured value F of the proposed algorithm is improved by 15%.

show abstract

Masked-RPCA: Moving Object Detection With an Overlaying Model

Cited by 6 publications

References 52 publications

Mainlobe jamming suppression via improved BSS method for rotated array radar

Mainlobe jamming suppression via improved BSS method for rotated array radar

Interpretable Neural Networks for Video Separation: Deep Unfolding RPCA with Foreground Masking

Moving Object Detection Based on Weighted Kernel Norm and Saliency Constrained RPCA

Contact Info

Product

Resources

About