Adaptive-Rate Compressive Sensing Using Side Information.

Area of Science:

• Signal Processing
• Computer Vision
• Applied Mathematics

Background:

• Compressive Sensing (CS) is an advanced signal acquisition technique.
• Sparse, time-varying signals present unique challenges for traditional CS methods.
• Existing CS techniques often require prior knowledge of signal sparsity, limiting their applicability.

Purpose of the Study:

• To develop novel adaptive-rate CS strategies for sparse, time-varying signals.
• To leverage side information for improved CS performance.
• To address limitations in current CS methods that assume a known sparsity bound.

Main Methods:

• Introduced two adaptive-rate CS strategies utilizing side information.
• Method 1: Employed additional cross-validation measurements.
• Method 2: Exploited additional low-resolution measurements.
• Developed techniques for background subtraction using spatially multiplexing CS cameras (e.g., single-pixel camera).
• Predicted the number of significant coefficients at the next time instant using side information, avoiding prior sparsity bound assumptions.

Main Results:

• Proposed techniques dynamically adjust the number of CS measurements per image in a video sequence.
• Experimental validation on real surveillance video sequences demonstrated the effectiveness of the proposed methods.
• Achieved efficient signal reconstruction for sparse, time-varying signals without assuming a fixed sparsity level.

Conclusions:

• The novel adaptive-rate CS strategies offer a more flexible and efficient approach to signal acquisition.
• Leveraging side information significantly enhances CS performance for dynamic scenes.
• These methods are particularly suitable for applications like background subtraction with CS cameras.