Adaptive Rate Block Compressive Sensing Based on Statistical Characteristics Estimation

In some video compressive sensing (CS) applications, the sparsity of original signals is unknown to the sampling device. The computing power, memory space and power consumption of the sampling device are also limited, which makes it difficult to achieve adaptive rate compressive sensing (ARCS). A new blocked ARCS method for surveillance videos is proposed, which fully considers the limitations mentioned above. By observing the result of CS measurement, the statistical characteristics of the original signal are estimated. The sparsity of the original signal is reasonably estimated by using these statistical characteristics. Therefore, blocks can be divided into more classes with higher accuracy. The proposed method has the advantages of low computational complexity, small memory footprint and low power consumption, which makes it suitable for implementing in applications such as wireless video sensor networks (WVSN) and single pixel cameras (SPC). The experiment results show that the proposed method can well adapt to the change of sparsity, allocate appropriate sampling rate for each block, effectively reduce the sampling rate, and improve the quality of the reconstructed image. Meanwhile, the amount of calculation in the sampling process is much lower, and the sampling speed is obviously accelerated. The overall performance of the proposed method is better than the previous state-of-the-art method.

Automated summary

This paper proposes a new blocked ARCS method for surveillance videos, which estimates the sparsity of original signals by observing the result of CS measurement, and achieves higher accuracy in block division. The method has the advantages of low computational complexity, small memory footprint, and low power consumption, making it suitable for various applications.