The invention relates to a compressed sensing video reconstruction method based on local and non-local constraints. A local constraint and a non-local constraint are introduced into a compressed sensing reconstruction video model, and the inherent characteristics of a three-dimensional video sequence are guided and optimized by utilizing a total variation (TV) constraint based on an image frame and a low-rank constraint based on a sequence time-space similar block matrix. In the optimization problem solving process, firstly, a traditional method is adopted to reconstruct each frame of image compressed sensing image in a video; an initial estimation of an obtained image frame is obtained, then an auxiliary variable is introduced, the problem is decomposed into a TV optimization sub-problemand a low-rank optimization sub-problem, solving is carried out alternately, and the TV sub-problem and the low-rank optimization sub-problem are further decomposed into a differentiable part and a solved part to be solved respectively in an internal loop. According to the method, the spatio-temporal local and non-local features of the image frame are fully considered, so that the reconstruction quality of the image is improved.