The invention discloses a non-convex optimization based MIMO radar moving object detection method. The method comprises the following steps: according to radar parameters, performing distance-angle-Doppler grid dividing on an observation area; according to radar emission, the position of a receiving array, and radar observation area parameters, calculating emission and reception guiding vectors; according to emission signal waveforms, the guiding vectors and a distance unit time delay transformation matrix, constructing a sparse dictionary matrix; performing serialization on echo signals received by an antenna array; and according to an aforementioned model, constructing MIMO radar moving object detection to be an optimization problem of an L[1/2] norm minimizing constraint; and using a heavy weight determining L[1] norm regularization method to solve the optimization problem, obtaining distance-angle-Doppler imaging of a MIMO radar moving object, and detecting a moving object in the observation area. By using the method provided by the invention, a moving object detection result more accurate than a result by use of a convex optimization algorithm can be obtained, and the detection result is higher in resolution.