The invention discloses a monitoring video-oriented foreground and background parallel compression method based on deep learning. A video foreground and background separation step; a self-adaptive background template updating step: judging whether to update the background template or not through comparison; an image frame partitioning and compressing step: partitioning, segmenting and compressingthe whole image frame; a two-order coarse-to-fine decoding step: sequentially inputting each block code obtained by partitioning and compressing the image frame into a composite network and a reconstruction network; constructing an adversarial network structure, and training in combination with compression and decoding steps; and performing parallel compression on each image frame of the monitoring video to be compressed by adopting the trained image frame block compression step. On the premise of ensuring the video compression quality, the video compression ratio is greatly improved, the real-time performance is better, the encoding and decoding time is shorter, and the application scene is wider.