Compressive coded video compression using measurement domain motion estimation

In this paper, we propose a novel CS video compression scheme having a simple encoder and a decoder that performs measurement domain motion estimation in order to reconstruct the video frames. In the encoder, the sensing mechanism can be applied directly on the non-sparse video frame using a random sensing matrix and the frame can be reconstructed at the decoder using l₁-minimization or any of the iterative greedy algorithms provided the sensing matrix satisfies D-RIP with a high probability. For reconstructing a video frame at the decoder, if the motion is estimated prior to reconstruction, CS recovery techniques can efficiently recover the sparse frame difference. In our previous work, we have shown that the motion estimation is possible in the measurements domain by using a circulant CS matrix instead of a random matrix. We propose to estimate the frame motion from the circulant CS measurements and reconstruct the video frame using the motion compensated frame difference sparsity. Experimental results show that our proposed method performs better than the traditional CS video compression schemes.