A video processing method includes dividing a current frame into a plurality of blocks, generating a motion vector of each block of the plurality of blocks of the current frame according to the each block of the current frame and a corresponding block of a previous frame, generating a global motion vector according to a plurality of motion vectors of the current frame, generating a sum of absolute differences of pixels of each block of the current frame according to the global motion vector, generating a region with a set of blocks of the current frame, matching a distribution of the sum of absolute differences of pixels of the region with a plurality of models, identifying a best matching model, and labeling each block in the region in the current frame with a label of either a foreground block or a background block according to the best matching model.