Flowchart of the proposed framework. First, the video BG is modeled by fuzzy factorization. Second, background subtraction is conducted, and then the Spatio-temporal constraints are applied to obtain the FG component. After that, the Gaussian noise in residual and the FG component update the membership degree together. Finally, the above process is iterated until convergence. ( DOI: 10.1007/s11704-022-2099-0)
Robustness to different noises. The source video frame and the result from the study’s JMDF are displayed in the lower right corner of the figure. The research team add Gaussian noise, speckle noise, salt and pepper noise, and Poisson noise to the source video, respectively. Except for Poisson noise, they fix the noise mean to 0, and change the noise variance, whose values are shown in the first row of each sub-figure. Then, the team plot the noisy frame examples in the second row and the results from JMDF in the third row. As the variance increases, the video gradually becomes blurred, which increases the difficulty of the detection task. This method achieves good performance even in videos with large noise variance( DOI: 10.1007/s11704-022-2099-0)
ADMM algorithm for JMDF-based model
Newswise — Researchers have developed the Joint Matrix Decomposition and Factorization (JMDF) framework to enhance the process of detecting moving targets in video streams. This technology uses a combination of established and novel methods to improve both the accuracy and robustness of video analysis in dynamic environments.
Linhao Li, the lead researcher from Hebei University of Technology, explains, "The JMDF framework extends traditional Matrix Decomposition (MD) and Matrix Factorization (MF) methods by incorporating elements of fuzzy factorization and adaptive temporal differencing. This approach helps in effectively distinguishing between background and foreground elements in complex video scenes."
The technology incorporates two key methods:
- Fuzzy Logic: This approach allows the system to better handle uncertainties in video data. By applying what is known as fuzzy factorization, the framework can more accurately determine what parts of the video are background and which are moving objects.
- Adaptive Constraints: The framework adjusts its analysis based on the movement in the video. This adaptability helps it maintain accuracy even when the scene changes rapidly or unexpectedly.
The computational efficiency of the system is optimized using a method called the Alternating Direction Method of Multipliers (ADMM), which ensures that the video analysis is both fast and accurate. Performance tests on various video datasets confirm that JMDF maintains its effectiveness under different environmental conditions.
Future developments will focus on incorporating deep learning techniques with JMDF to further enhance its capabilities. "Our aim is to develop a system that adapts in real-time to changing environments, improving reliability in video analysis," says Li.
This technology has potential applications beyond conventional surveillance, including in autonomous vehicle navigation and traffic management systems, where accurate and real-time video analysis is essential.
The research, carried out in collaboration with West Virginia University, has been published in the journal Frontiers of Computer Science and is available via DOI: 10.1007/s11704-022-2099-0.
MEDIA CONTACT
Register for reporter access to contact detailsArticle Multimedia
Credit: Linhao Li
Caption: Flowchart of the proposed framework. First, the video BG is modeled by fuzzy factorization. Second, background subtraction is conducted, and then the Spatio-temporal constraints are applied to obtain the FG component. After that, the Gaussian noise in residual and the FG component update the membership degree together. Finally, the above process is iterated until convergence. ( DOI: 10.1007/s11704-022-2099-0)
Credit: Linhao Li
Caption: Robustness to different noises. The source video frame and the result from the study’s JMDF are displayed in the lower right corner of the figure. The research team add Gaussian noise, speckle noise, salt and pepper noise, and Poisson noise to the source video, respectively. Except for Poisson noise, they fix the noise mean to 0, and change the noise variance, whose values are shown in the first row of each sub-figure. Then, the team plot the noisy frame examples in the second row and the results from JMDF in the third row. As the variance increases, the video gradually becomes blurred, which increases the difficulty of the detection task. This method achieves good performance even in videos with large noise variance( DOI: 10.1007/s11704-022-2099-0)
Credit: Linhao Li
Caption: ADMM algorithm for JMDF-based model
RSS