Video Compressive Sensing

Video compressive sensing (VCS) aims to reconstruct high-speed videos from significantly undersampled measurements, enabling efficient video capture with low-frame-rate sensors. Current research focuses on developing efficient and robust deep learning-based reconstruction methods, often employing U-Net architectures or graph neural networks to leverage spatial and temporal correlations within video frames, and incorporating techniques like deep unfolding to improve both accuracy and speed. These advancements are driving progress towards real-time VCS on mobile devices, with potential applications in various fields requiring high-speed imaging, such as surveillance, sports analysis, and medical imaging.