A Tensor-Based Online RPCA Model for Compressive Background Subtraction.

Area of Science:

• Computer Vision
• Video Processing
• Machine Learning

Background:

• Background subtraction is crucial in computer vision.
• Online compressive measurements offer efficiency but face limitations.
• Existing matrix-based methods fail to preserve spatial structure and handle background disturbances.

Purpose of the Study:

• To propose a novel tensor-based method for online compressive video reconstruction and background subtraction.
• To address limitations of existing matrix-based methods.
• To enhance robustness in complex video scenes.

Main Methods:

• A tensor-based online compressive video reconstruction and background subtraction method (NIOTenRPCA) is proposed.
• The method explicitly models background disturbances as nonidentical but correlated noise.
• This approach preserves spatial structure and accounts for complex background variations.

Main Results:

• NIOTenRPCA demonstrates superior performance compared to existing state-of-the-art methods.
• Extensive experiments on real-world video datasets validate the method's effectiveness.
• The proposed method shows improved robustness in complex video scenes.

Conclusions:

• The proposed tensor-based approach offers a more robust solution for background subtraction from online compressive measurements.
• NIOTenRPCA effectively handles complex background disturbances, outperforming traditional matrix-based methods.
• The method advances the field of efficient video analysis and background subtraction.