Motion compensated compressive video sensing

In order to meet the requirement for low-complexity video encoding, an efficient compressive video sensing algorithm is proposed. Based on the compressive sensing (CS) theory, a low-complexity encoder directly captures compressed video data called measurements via randomly projecting. However, a high-complexity decoder performs reconstruction of CS frames by exploiting correlations among successive frames via motion compensated prediction and utilizing sparse recovery of prediction residual. Experimental results show that, our method may achieve up to 4dB gain in PSNR over the existing compressive video sensing algorithms for the same number of measurements, and is a promising candidate for wireless visual surveillance and visual sensor networks.