Total Variation Regularized Tensor RPCA for Background Subtraction From Compressive Measurements.

Area of Science:

• Computer Vision
• Video Analysis
• Machine Learning

Background:

• Background subtraction is crucial for video analysis tasks like surveillance.
• Compressive imaging has led to background subtraction from compressive measurements (BSCM) research.
• Existing methods may struggle with complex spatio-temporal correlations in video data.

Purpose of the Study:

• To propose a novel tensor-based robust principal component analysis (TenRPCA) for background subtraction from compressive measurements (BSCM).
• To enhance spatio-temporal continuity of foregrounds and model background correlations using tensor decomposition.
• To develop efficient algorithms for solving the proposed TenRPCA models.

Main Methods:

• Decomposition of video frames into background and foreground components within a tensor framework.
• Utilizing 3D total variation for foreground spatio-temporal continuity.
• Employing Tucker decomposition for background spatio-temporal correlations.
• Developing holistic and patch-group-based TenRPCA models.
• Implementing alternating direction method of multipliers for efficient algorithm development.

Main Results:

• The proposed holistic TenRPCA model effectively decomposes video frames.
• The patch-group-based TenRPCA model further characterizes background correlations.
• Efficient algorithms were successfully developed using the alternating direction method of multipliers.
• Extensive experiments demonstrated superior performance compared to state-of-the-art methods on simulated and real-world videos.

Conclusions:

• The novel TenRPCA approach offers a robust solution for background subtraction from compressive measurements.
• The tensor-based framework effectively models complex spatio-temporal structures in video data.
• The proposed methods show significant improvements in background subtraction accuracy and efficiency for video surveillance applications.