Dual-scale dual-rate video compressive sensing for edge surveillance devices

Classic video compression method suffers from long encode time and requires large memories, making it hard to deploy on edge devices, thus video compressive sensing which requires less resources during encoding is getting more attention. We propose a dual-scale dual-rate video compressive sensing algorithm for surveillance video compression. Proposed method extracts and compresses foreground area and reference frame separately using dual-scale compressive sampling, then using reversible neural network to reconstruct original frames. Finally we test compressive sampling and region of interest (ROI) extraction network in proposed method on edge device and reconstruction network on server. The experiments show that proposed method can fast compresses frame and extracts foreground area on edge computing devices, achieves higher reconstruction quality.