Adaptive video compressed sampling in the wavelet domain

• A multiscale framework for compressed video acquisition is proposed.
• Redundancy is removed by using wavelet-based sparse sampling and motion estimation.
• Video sequences can be retrieved from compressed measurements in real time.
• The proposed method allows adaptive control over the reconstructed video quality.
• The reconstruction time and measurements required can be significantly reduced.

In this work, we propose a multiscale video acquisition framework called adaptive video compressed sampling (AVCS) that involves sparse sampling and motion estimation in the wavelet domain. Implementing a combination of a binary DMD and a single-pixel detector, AVCS acquires successively finer resolution sparse wavelet representations in moving regions directly based on extended wavelet trees, and alternately uses these representations to estimate the motion in the wavelet domain. Then, we can remove the spatial and temporal redundancies and provide a method to reconstruct video sequences from compressed measurements in real time. In addition, the proposed method allows adaptive control over the reconstructed video quality. The numerical simulation and experimental results indicate that AVCS performs better than the conventional CS-based methods at the same sampling rate even under the influence of noise, and the reconstruction time and measurements required can be significantly reduced.