Abstract: Existing video surveillance implementations often use traditional video coding methods, which have high requirements on equipment cost, deployment environment and data storage due to high coding complexity. Compression sensing has great potential in solving problems related to video surveillance due to its natural distributed nature. However, there are still some difficulties in realizing adaptive rate compression sensing when the sparsity of the original video signal is unknown. In order to realize adaptive rate compression sensing in practical surveillance video applications, a time-domain adaptive rate sampling method and a time-frequency domain adaptive rate sampling method based on sub-block statistical characteristic estimation are proposed. The video is divided into appropriate sub-blocks, and the similar sub-blocks are quickly clustered by statistical characteristic estimation and clustering algorithm, and further the sampling rate is reasonably assigned to each type of sub-blocks to realize the adaptive rate compression sensing, which reduces the total sampling rate without degrading the reconstruction quality and significantly increasing the computational complexity of sampling. Simulation results show that, compared with the existing video adaptive rate compression sensing scheme, the proposed method can more reasonably allocate the sampling rate for sub-blocks with different sparsity and reduce the total sampling rate, and at the same time the sampling computational complexity is not high, which can be accepted by the actual sampling equipment.

Key words: Surveillance video, Compressed sensing, Adaptive rate sampling, Statistical characteristic estimation, sub-block clustering